FPCD: An Open Aerial VHR Dataset for Farm Pond Change Detection

Chintan Tundia

∗ a

, Rajiv Kumar

∗ b

, Om Damani

and G. Sivakumar

Indian Institute of Technology Bombay, Mumbai, India

Keywords:

Object Detection, Instance Segmentation, Change Detection, Remote Sensing.

Abstract:

Change detection for aerial imagery involves locating and identifying changes associated with the areas of in-

terest between co-registered bi-temporal or multi-temporal images of a geographical location. Farm ponds

are man-made structures belonging to the category of minor irrigation structures used to collect surface

run-off water for future irrigation purposes. Detection of farm ponds from aerial imagery and their evo-

lution over time helps in land surveying to analyze the agricultural shifts, policy implementation, seasonal

effects and climate changes. In this paper, we introduce a publicly available object detection and instance

segmentation (OD/IS) dataset for localizing farm ponds from aerial imagery. We also collected and anno-

tated the bi-temporal data over a time-span of 14 years across 17 villages, resulting in a binary change de-

tection dataset called Farm Pond Change Detection Dataset (FPCD). We have benchmarked and analyzed

the performance of various object detection and instance segmentation methods on our OD/IS dataset and

the change detection methods over the FPCD dataset. The datasets are publicly accessible at this page:

https://huggingface.co/datasets/ctundia/FPCD.

1 INTRODUCTION

Accurate and timely detection of geographical

changes on earth’s surface gives extensive informa-

tion about the various activities and phenomena hap-

pening on earth. Change detection task helps in an-

alyzing and understanding co-registered images for

change information. A change instance between two

images refers to the semantic level differences in the

appearance between the two images in association

to the regions of interest captured at different points

in time. On geographical images, it helps to keep

track of the changes to analyze evolution of land ge-

ography or land objects and to mitigate hazards at

local and global scales. The availability of high-

resolution aerial imagery has enabled land use and

land cover monitoring to detect objects such as wells,

farm ponds, check dams, etc. at instance level.

Change detection can be bi-temporal when two

points in time are compared or multi-temporal when

multiple points in time are compared. When multi-

temporal data is captured by satellites, drones or aerial

https://orcid.org/0000-0003-3169-1775

https://orcid.org/0000-0003-4174-8587

https://orcid.org/0000-0002-4043-9806

https://orcid.org/0000-0003-2890-6421

∗

These Authors made equal contribution

vehicles, they are constrained by spatial, spectral and

temporal elements, in addition to atmospheric condi-

tions, resolution, etc. The cloud cover, shadows and

seasonal changes become visible in many aerial im-

ages that affect the overall appearance of the images

resulting in the co-registered images to appear from

different domains. Moreover, acquiring and annotat-

ing images from satellite images to build change de-

tection datasets is a costly process that involves many

underlying tasks. The absence of paired images of

the same location captured at different times makes it

difﬁcult to obtain a useful change detection dataset.

Also when paired bi-temporal images are available,

there might not be changes present, or the bi-temporal

images are captured in very short intervals.

Figure 1: Different categories of farm ponds. From left to

right: Wet farm pond (lined), Dry farm pond (lined), Wet

farm pond (unlined) and Dry farm pond (unlined).

Farm ponds have become popular as private irri-

gation sources in developing countries like India over

the last two decades. A farm pond is an artiﬁcial dug

out structure having an inlet and outlet for collect-

862

Tundia, C., Kumar, R., Damani, O. and Sivakumar, G.

FPCD: An Open Aerial VHR Dataset for Farm Pond Change Detection.

DOI: 10.5220/0011797600003417

In Proceedings of the 18th International Joint Conference on Computer Vision, Imaging and Computer Graphics Theory and Applications (VISIGRAPP 2023) - Volume 5: VISAPP, pages

862-869

ISBN: 978-989-758-634-7; ISSN: 2184-4321

 2023 by SCITEPRESS – Science and Technology Publications, Lda. Under CC license (CC BY-NC-ND 4.0)

ing the surface runoff water ﬂowing from the farm

area (Tundia et al., 2020). They are used to col-

lect and store rainwater so as to provide irrigation to

crops during periods of water scarcity. It is one of

the many minor irrigation structures (Tundia. et al.,

2022) with a cultivable command area of up to 2000

hectares. Farm ponds can be classiﬁed into two cate-

gories based on the presence or absence of water in it:

wet farm ponds and dry farm ponds. Also, farm ponds

can be either lined or unlined depending on the use of

plastic lining to prevent water seepage into ground-

water. Overall, farm ponds can be classiﬁed into 4

sub-categories (See Fig 1): lined wet farm pond, un-

lined wet farm pond, lined dry farm pond and unlined

dry farm pond based on their structure and the pres-

ence of water.

Though efforts have been put up to promote farm

ponds, serious concerns have been raised over their

implementation and their usage. The purpose of

building farm ponds has long drifted from their orig-

inal objective of storing rainwater for protective ir-

rigation to being used as storage tanks for pumped-

out groundwater exposing the underground water to

evaporation losses. In the due time, farm ponds have

accelerated the rate of groundwater exploitation to

many folds (Prasad et al., 2022). The objective of

our work is to develop change detection models to de-

tect and visualize changes associated with farm ponds

and to compute percentage differences in the pres-

ence/absence of wet/dry farm ponds. This can help

in making policy decisions and in analyzing impacts

of farm ponds on aspects like shifting of agricultural

practice, policies being implemented, seasonal effects

and changes, etc.

1.1 Contributions

Our contributions in this paper are:

1. Farm pond change detection dataset (FPCD):

A publicly available dataset for change detection

tasks on farm pond categories for different pur-

poses and stakeholders.

2. A small-scale public dataset for object detection

and instance segmentation of four farm pond cat-

egories.

The paper is organized as follows: section 1.3 cov-

ers the related work, section 2 covers the details of the

proposed dataset, section 3 covers the experiments,

section 4 covers the results and observation, and ﬁ-

nally conclusion in section 5.

Table 1: Comparison of change detection datasets.

Dataset

Image

Pairs

Res.

(m)

Image

Size

Bands Mask Type

Change

Inst.

CLCD

(Liu et al., 2022)

600 0.5 to 2 512 x 512 RGB Binary -

S2Looking

(Shen et al., 2021)

5000

0.5

0.8

1024 x 1024 RGB Binary 65,920

SYSU-CD

(Shi et al., 2022)

20000 0.5 256 x 256 RGB Binary -

DSIFN

(Zhang et al., 2020)

3988 - 512 x 512 RGB Binary -

LEVIR-CD

(Chen and Shi, 2020)

637 0.5 1024 x 1024 RGB Binary 31,333

WHU

Building

CD (Ji et al., 2019)

1 0.075

32207

15354

RGB Binary 2297

SZTAKI

(Bourdis et al., 2011)

13 1.5 952 x 640 RGB Binary 382

FPCD 694 0.156 1024 x 768 RGB

Binary

616

1.2 Problem Formulation

Generally, change detection tasks involve an input

set of multi-temporal images and the correspond-

ing ground truth mask, with most change detection

datasets having bi-temporal images mapped to binary

mask labels. A general assumption in most change

detection tasks is that there is a pixel-to-pixel corre-

spondence between the two images and these corre-

spondences are registered to the same point on a ge-

ographical area. Based on the correspondences, each

pixel in the change mask can be assigned a label in-

dicating whether there is a change or not. In other

words, pixels belonging to the change mask are as-

signed a change label if the corresponding area of in-

terest has geographical changes with respect to each

other and those are not assigned a change label when

there are no changes. In a binary change detection

setting, the paired input images are T

∈ R

CxHxW

and

∈ R

CxHxW

with a size of CxHxW, where H and W

are the spatial dimensions and C=3 is the input image

channel dimensions. The ground truth mask can be

represented as pixel based mask label, M ∈ R

CxHxW

for the bi-temporal input images of the change detec-

tion task.

1.3 Related Work

1.3.1 Datasets

Change detection datasets generally use RGB images

(Chen and Shi, 2020) or hyperspectral images (Daudt

et al., 2018) with some datasets (Van Etten et al.,

2021) having change instances up to 11 million. The

image resolution of CD datasets range from a few

centimeters (Ji et al., 2019), (Shao et al., 2021), (Tian

et al., 2020) to 10 meters (Van Etten et al., 2021),

(Daudt et al., 2018). Some of these datasets have

image pairs ranging from a few hundred (Liu et al.,

2022) to tens of thousands (Shi et al., 2022) and some

have very few input image pairs of very high reso-

lution. While most datasets have ﬁxed image sizes.

FPCD: An Open Aerial VHR Dataset for Farm Pond Change Detection

863

Most of the CD datasets have modest image sizes ex-

cept for the exception of a few (Wu et al., 2017), (Ji

et al., 2019). We summarize the different aspects of

various change detection datasets along with their de-

tails in Table 1.

1.3.2 Change Detection Techniques

Deep learning based change detection methods can be

classiﬁed into feature-based, patch-based and image

based deep learning change detection. Earlier CNN

architectures used siamese with triplet loss (Zhang

et al., 2019) and weighted contrastive loss (Zhan

et al., 2017) to learn discriminative features between

change and no-change images, and used Euclidean

distances between the images features to generate

the difference images. Recent developments in deep

learning have led to the usage of attention mecha-

nisms and feature fusion at various scales improving

feature extraction capabilities. We compare and list

some of the existing encoder-decoder based models

used for change detection below.

Deeplabv3+ (Chen et al., 2018) uses spatial pyra-

mid pooling module to encode multi-scale context at

multiple effective ﬁelds-of-view and encode-decoder

structure to capture sharper object boundaries. It im-

proves upon DeepLabv3 (Chen et al., 2017) by apply-

ing depthwise separable convolution to both Atrous

Spatial Pyramid Pooling and decoder modules. Pyra-

mid scene parsing network (PSPNet)(Zhao et al.,

2017) along with pyramid pooling module applies

different-region-based context aggregation to pro-

duce global prior representation for pixel-level pre-

diction tasks. Uniﬁed Perceptual Parsing Net (UPer-

Net)(Xiao et al., 2018) is a multi-task framework that

can recognize visual concepts from a given image us-

ing a training strategy developed to learn from het-

erogeneous image annotations. Multi-scale Attention

Net (MA-Net) (Fan et al., 2020) uses multi-scale fea-

ture fusion to improve the segmentation performance

by introducing self-attention mechanism to integrate

local features with their global dependencies. It uses

Position-wise Attention Block (PAB) to model the

feature inter-dependencies in spatial dimensions and

Multi-scale Fusion Attention Block (MFAB) to cap-

ture the channel dependencies between any feature

map by multi-scale semantic feature fusion.

LinkNet (Chaurasia and Culurciello, 2017) pro-

poses a novel deep neural network architecture that

uses only 11.5 million parameters learning with-

out any signiﬁcant increase in number of parame-

ters. Pyramid Attention Network (PAN) (Li et al.,

2018) uses Feature Pyramid Attention module and

Global Attention Upsample module to combine at-

tention mechanism and spatial pyramid to extract

precise dense features for pixel labeling. Unet++

(Zhou et al., 2018) improves on Unet (Ronneberger

et al., 2015) with new skip pathways and is based

on the idea that optimizer can learning easily with

semantically similar feature maps by reducing the

gap between the feature maps of the encoder and de-

coder sub-networks. Bitemporal image transformer

(BiT) (Hao Chen and Shi, 2021) models contexts

within the spatial-temporal domain in a deep feature

differencing-based CD framework. The bi-temporal

images are encoded as tokens and a transformer en-

coder models the contexts in the compact token-based

space-time. A transformer decoder then reﬁnes the

original features from the learned context-rich tokens

back to the pixel-space.

1.3.3 Object Detection and Instance

Segmentation Techniques

Object detection is a computer vision task that local-

izes and classiﬁes objects of interest in an image. In-

stance segmentation on the other hand provides a de-

tailed inference for every single pixel in input image.

Since the advent of deep learning(Kumar et al., 2021),

object detection has vastly beneﬁted from sophisti-

cated architectures and image representations. Over

the years, deep learning based detectors have evolved

into one-stage and two-stage detectors. In one-stage

detection, the input image is divided into regions si-

multaneously with the probabilistic prediction of ob-

jects, while in two-stage object detection the object

proposals are classiﬁed in the second stage from a

sparse set of candidate object proposals generated in

the ﬁrst stage. A few of the well-known two-stage de-

tectors include the FasterRCNN (Ren et al., 2017),

GridRCNN (Lu et al., 2019), while the YOLOv3

(Redmon and Farhadi, 2018), Generalized Focal Loss

(Li et al., 2020), Gradient Harmonized SSD (Li et al.,

2019) are one-stage detectors. Detecting objects from

aerial imagery is affected by adversities like view-

point variation, illumination, occlusion, etc. and be-

comes difﬁcult due to objects being small, sparse and

non-uniform in high-resolution aerial images.

2 FARM POND CHANGE

DETECTION (FPCD) DATASET

FPCD is a novel publicly available change detection

dataset that focuses on changes associated with irri-

gation structures from India. The details of the multi-

temporal images including the district, village name

and the time-interval over which the images were col-

lected are summarized in Table 5.

VISAPP 2023 - 18th International Conference on Computer Vision Theory and Applications

864

Table 2: Temporal object instances and their respective

change classes. Row 1 corresponds to T0 images, row 2 cor-

responds to T1 images, row 3 corresponds to change mask.

Columns - (a,b) No Change, (c) Farm pond constructed, (d)

Farm pond demolished, (e) Farm pond dried and (f) Farm

pond wetted.

a b c d e f

Table 3: Class distribution of change objects.

Change

Class

No. of

Instances

Farm pond constructed 431

Farm pond demolished 39

Farm pond dried 47

Farm pond wetted 99

2.1 Dataset Collection Technique

The images were collected by following the steps

given below:-

1. Area and Timestamp Selection: The ground

truth information about the farm pond locations

and the list of villages were collected from dif-

ferent sources like news articles, reports and web-

pages along with Jalyukt Shivar data(Maharashtra

Remote Sensing Application Centre (MRSAC),

2015). The location of the Farm ponds would

span across the different districts of Maharashtra,

India. The villages were then rigorously ﬁltered

and chosen based on the presence or absence of

farm ponds, and bi-temporal pairs were decided

by visual inspection of various timestamps.

2. Grid Formation: For collecting the images of a

ﬁxed size for each village, the grids of latitude

and longitude at a ﬁxed zoom level of Google

Maps were needed. We selected zoom level as

18 which provides resolution of sub-meter level.

Grids of size 1024x768 pixels were created by

using the village boundaries provided by Indian

Village Boundaries Project (Data{Meet}, 2019)

in the geo-location dimensions i.e. latitudes and

longitudes.

3. Image Collection: Google Earth Pro desktop

software was used for collecting historical im-

agery. For each grid cell, the map was set such

that it would cover the grid cell boundaries with

the view. Then the bi-temporal images were saved

by setting the timestamp one at a time using the

Google Earth Historical Imagery tool.

Table 4: Class distribution of object instances in OD/IS

dataset.

Annotation

Class

No. of

Instances

Wet farm pond (lined) 287

Wet farm pond (unlined) 293

Dry farm pond (lined) 90

Dry farm pond (unlined) 668

Table 5: Villages and timestamps for FPCD dataset.

District Village Name Timestamp

#Image

Pairs

Nasik Hadap Sawargaon Feb-2019, Feb-2021 25

Akola Akhatwada Mar-2007, Mar-2018 24

Akola Ghusar April-2013, Mar-2018 124

Amravati Nardoda Nov-2009, Jan-2021 39

Aurangabad Kumbephal Feb-2014, Mar-2020 62

Beed Kumbhephal April-2012, Mar-2019 37

Buldhana Bhivgaon Jan-2014, Jan-2019 8

Hingoli Gondala Feb-2014, Mar-2020 46

Jalgaon Pimpalgaon Block Oct-2013, Nov-2020 17

Jalna

Dawargaon/

Bhatkheda

April-2012, April-2019 16

Latur Sumthana Jan-2013, Jan-2021 15

Nanded Chainpur Jan-2014, Mar-2019 56

Osmanabad Ambejawalga Jan-2014, Feb-2019 74

Parbhani Kinhola Block May-2010, April-2020 62

Wardha Kakaddara Jan-2013, Feb-2019 17

Washim Bhoyata April-2013, Feb-2021 46

Yavatmal Wadhona Pilki April-2014, April-2019 26

Total Images 694

4. Object & Instance Annotation: Once we have

collected all the image pairs for each village, we

annotate the objects individually in each image.

Depending on the types of farm ponds, there are

four main classes as given in Table 4. We use the

annotation tool, LabelMe(Russell et al., 2008) to

annotate and label the above mentioned classes.

And further we convert them into CoCo(Lin et al.,

2014) format required for the object detection

and instance segmentation tasks thus forming the

Farm pond OD/IS dataset.

5. Change Mask Generation: In this step, we

take bi-temporal pair of images and correspond-

ing farm pond annotations. Based on the loca-

tion and farm pond category we generate differ-

ent types of change masks with change classes

as given in Table 2. More details on the type of

changes are given in subsection 2.2.

2.2 Dataset Details

A total of 694 images of size 1024 x 768 pixels at

zoom level 18 were collected from Google Earth im-

ages using the technique described in subsection 2.1.

The regions of Maharashtra in India were chosen,

since it is largely a groundwater dependent region in

western India. The images collected at zoom level

18 are at a very high resolution scale, up to 1 me-

ter. The details of villages collected and their respec-

tive timestamps are given in Table 5. Most of the

villages have timestamps during the months of Jan-

April and the minimum year difference between bi-

temporal images is 2 years and the maximum year

difference is 9 years, earliest being 2007 and latest be-

ing 2021. The FPCD dataset consists of image pairs,

FPCD: An Open Aerial VHR Dataset for Farm Pond Change Detection

865

change masks and object annotations of farm ponds

as polygons set in COCO (Lin et al., 2014) format.

For farm pond change detection, we identify four

change classes i.e Farm pond constructed, Farm pond

demolished, Farm pond dried and Farm pond wetted.

Let’s consider the bi-temporal pair as T0-T1, T0 be-

ing the image captured with the old timestamp and

T1 being the image captured with new timestamp im-

age. We identify a binary change as FP constructed,

when there is no farm pond in the T0 image, but a

farm pond is observed at the same location in the T1

image. Likewise, we identify a binary change as FP

demolished, when a farm pond existed at any location

in the T0 image and there is no farm pond at the same

location or replaced by different terrain in the T1 im-

age. We identify a binary change as farm pond dried

when there existed a wet farm pond in the T0 image

and the farm pond at the same location in T1 image

becomes dry due to absence of water or presence of

low water level. We identify a binary change as farm

pond wetted, when there existed a dry farm pond in

the T0 image and the farm pond at the same location

in T1 image is ﬁlled with water or the water reaches

the surface level.

Some farmers may pump groundwater and ﬁll

the farm ponds instead of using surface-water runoff

due to rain leading to water loss by further evapo-

ration. This leads to depletion of groundwater and

the practice of farm pond based irrigation becom-

ing unsustainable (Prasad et al., 2022). Thus group-

ing farm pond constructed and farm pond demolished

class helps in identifying the increase/decrease in

farm ponds. This helps stakeholders like researchers

and policy makers to monitor the impacts due to in-

crease/decrease in farm ponds like change in agricul-

tural patterns, impact on ground water, etc. We clas-

sify this task as Task-1.

In certain dry and semi-arid regions, due to un-

certain climatic conditions, yearly and seasonal rain-

fall is erratic. Agricultural ofﬁcers and researchers

often correlate climatic conditions and ground water

levels with such changes to infer further observations

eventually leading to necessary interventions needed

for agricultural sustainability. Thus we group Farm

pond dried and Farm pond wetted into Task-2. We

also combined the above grouped changes so as to

also provide overall statistics. We call this as Task-

3. We further explain experimental details of Task-1,

Task-2 and Task-3 in Section 3. The details of change

class distribution are given in the Table 3.

Table 6: Distribution of positives and negative image pairs

of various change detection tasks.

Task 1 Task 2 Task 3

Type Neg Pos Neg Pos Neg Pos

Train 363 237 534 66 329 271

Validation 52 42 83 11 45 49

Table 7: Binary change detection benchmark on farm pond

constructed/demolished task (Task 1).

ResNet - 18 ResNet - 50 ResNet - 101

Model Precision Recall F-Score Precision Recall F-Score Precision Recall F-Score

UNet 0.8174 0.8537 0.7360 0.9331 0.5533 0.5534 0.9051 0.7072 0.666

UNet++ 0.8467 0.8439 0.7668 0.8465 0.8439 0.7668 0.9123 0.7969 0.7669

MANet 0.8347 0.5536 0.5539 0.9998 0.5531 0.5532 0.9894 0.5596 0.5539

LinkNet 0.8558 0.8251 0.7529 0.8580 0.7991 0.7226 0.9029 0.7554 0.7233

FPN 0.8833 0.8106 0.7567 0.8559 0.8446 0.7788 0.8582 0.8764 0.8003

PSPNet 0.9763 0.5855 0.5840 0.9763 0.5855 0.5840 0.9496 0.6178 0.6172

PAN 0.8572 0.8037 0.7250 0.8546 0.8383 0.7591 0.9176 0.7907 0.7685

DeepLabV3 0.8474 0.8428 0.7776 0.9138 0.7917 0.7720 0.8891 0.8188 0.7894

DeepLabV3+ 0.8998 0.8085 0.7870 0.8953 0.8151 0.7717 0.8563 0.8348 0.7771

UPerNet 0.8843 0.8148 0.7783 0.9055 0.8209 0.7855 0.9131 0.8389 0.8067

BiT 0.9252 0.8283 0.8704 - - - - - -

3 EXPERIMENTS AND

EVALUATION

Tasks like instance segmentation (Lin et al., 2014) and

image segmentation (Lin et al., 2014) involves a sim-

ilar pipeline like that of change detection. We used

the change detection framework (Kaiyu Li, 2021) for

conducting most of our CD-based experiments. The

pytorch based change detection framework supports

many models, encoders and deep architectures. In

this change detection pipeline, the bi-temporal images

are encoded into feature vectors using two encoders.

The encoders can be either siamese or non-siamese

type. The input images are encoded into encoded fea-

ture vectors, which are fused either by concatenating,

summing, subtracting or ﬁnding the absolute differ-

ence of the vectors. The resulting feature vector are

decoded as output which is compared to the ground

truth mask. We apply simple transforms like ﬂipping,

scaling and cropping for augmentations of the dataset

images at the global image level. Unlike other tasks,

for change detection the same transform has to be ap-

plied to the multi-temporal images.

3.1 Evaluation Criteria

We report the precision, recall and F-score values as

the metric to compare the performance in different

change detection tasks under various encoder back-

bones settings which is deﬁned as below:

Precision =

T P

T P + FP

, Recall =

T P

T P + FN

F-Score =

2 X Precision X Recall

Precision + Recall

where, TP is the number of true positives, FP is the

number of false positives, TN is the number of true

negatives and FN is the number of false negatives.

We report the bounding box mean average precision

VISAPP 2023 - 18th International Conference on Computer Vision Theory and Applications

866

Table 8: Binary change detection benchmark on farm pond

dried/wetted task (Task 2).

ResNet - 18 ResNet - 50 ResNet - 101

Model Precision Recall F-Score Precision Recall F-Score Precision Recall F-Score

UNet 0.9973 0.9094 0.9171 0.9893 0.9091 0.9082 0.9907 0.9128 0.9131

UNet++ 0.9813 0.9188 0.9112 0.978 0.9181 0.90573 0.9785 0.9147 0.9065

MANet 0.9826 0.9075 0.897 0.9927 0.9023 0.9052 0.9927 0.8966 0.902

LinkNet 0.9935 0.8867 0.8893 0.994 0.8974 0.9014 0.9918 0.91 0.9119

FPN 0.9923 0.9057 0.9063 0.9936 0.9125 0.9125 0.9949 0.9125 0.9133

PSPNet 0.9939 0.9054 0.907 0.9948 0.9023 0.9052 0.9926 0.9054 0.9064

PAN 0.9907 0.9146 0.9116 0.9905 0.9132 0.9113 0.9812 0.9135 0.901

DeepLabV3 0.9814 0.9066 0.8971 0.9326 0.9246 0.8698 0.9687 0.9096 0.8888

DeepLabV3+ 0.9873 0.9081 0.908 0.9918 0.9126 0.9109 0.9767 0.9246 0.912

UPerNet 0.9936 0.91367 0.9147 0.9957 0.9115 0.911 0.9929 0.9106 0.9096

BiT 0.8458 0.8254 0.8352 - - - - - -

Table 9: Comparison between results of training with all

image pairs and training with only positives images on Task

With all image pairs With only positive image pairs

ResNet - 18 ResNet - 50 ResNet - 18 ResNet - 50

Model Precision Recall F-Score Precision Recall F-Score Precision Recall F-Score Precision Recall F-Score

UNet 0.8627 0.7778 0.7444 0.864 0.7794 0.7442 0.798 0.7933 0.7049 0.8163 0.7967 0.7253

UNet++ 0.8961 0.7795 0.7579 0.8653 0.8081 0.7575 0.7884 0.8008 0.7013 0.8441 0.8482 0.768

MANet 0.7868 0.821 0.741 0.8451 0.7586 0.7246 0.7115 0.7925 0.6388 0.7944 0.7054 0.6145

LinkNet 0.8592 0.7872 0.7268 0.7868 0.8604 0.7502 0.7667 0.8024 0.6871 0.8753 0.7627 0.7236

FPN 0.8775 0.7856 0.7653 0.9192 0.7591 0.7696 0.9067 0.6843 0.6909 0.8403 0.8094 0.7442

PSPNet 0.8749 0.5652 0.5519 0.9571 0.5767 0.5919 0.8518 0.589 0.5564 0.8783 0.6212 0.601

PAN 0.8804 0.7252 0.7048 0.7228 0.6935 0.5435 0.752 0.7977 0.6575 0.6464 0.7536 0.5753

DeepLabV3 0.9329 0.7462 0.7652 0.8775 0.8257 0.7952 0.7656 0.8276 0.6813 0.7744 0.8611 0.7169

DeepLabV3+ 0.8232 0.7953 0.7238 0.9522 0.7118 0.7519 0.8046 0.8214 0.7315 0.8168 0.8045 0.7296

UPerNet 0.8634 0.7801 0.7515 0.9198 0.773 0.7704 0.8414 0.8043 0.7569 0.8473 0.8103 0.7575

BiT 0.8840 0.8647 0.8741 - - - 0.8710 0.8758 0.8734 - - -

(bbox mAP) and segmentation mean average preci-

sion (segm mAP) as the performance metric to com-

pare the various object detection and instance seg-

mentation methods under different backbone settings

in the benchmark. Intersection Over Union (IOU) is

a measure that evaluates the overlap between ground

truth and predicted bounding boxes and helps to de-

termine if a detection is True Positive. Average Pre-

cision (AP) is obtained by interpolating the precision

at each recall level r, taking the maximum precision

whose recall value is greater than or equal to r + 1.

Finally the mean of AP of all classes gives mean Av-

erage Precision (mAP), the metric being used to com-

pare different detectors.

3.2 Experiments

We benchmarked various object detection (OD) and

instance segmentation (IS) methods on the farm pond

OD/IS dataset. We use an existing encoder-decoder

based change detection framework (Kaiyu Li, 2021)

for the binary change detection tasks corresponding

to Farm pond construction/ demolition (Task 1) and

Farm pond dried/ wetted (Task 2). The encoder be-

longs to siamese type and the features from each

branch are fused by concatenation. A few experi-

ments were conducted to analyze the effect of vari-

ous components of the change detection pipeline. We

combined the masks from Task 1 and Task 2 to under-

stand the impact of performance on multiple change

classes. Combining the tasks lead to masks having

more change objects than either of the tasks (Refer

Fig 2). The train-validation split for each Task has

been mentioned in Table 6 . We split the farm ponds

dataset for OD/IS task in a 80/20 train/test proportion.

We used the MMDetection (Chen et al., 2019) frame-

work for implementing the various methods in the

Figure 2: T0 and T1 input image pair in the top row and the

image masks correspondingly for Task 1, Task 2 and Task 3

from left to right in the bottom row.

OD/IS benchmark. We used the pretrained backbones

of ResNet-50, ResNet-101 and MobileNetv2 primar-

ily trained on Imagenet, COCO and cityscapes for the

object detection experiments and Resnet-50, Resnet-

101 and Swin Transformer pretrained backbones for

instance segmentation experiments.

4 RESULTS AND OBSERVATIONS

The results of change detection on Farm pond

construction/demolition (Task-1) taking ResNet-18,

ResNet-50 and ResNet-101 as encoders are given in

Table 7. BiT model despite not having best precision

or recall values performs the best in terms of F-score

with the score of 0.8704. MANet despite perform-

ing highest in precision scores poorly in recall values

leading to poor F-score and FPN scores the highest

recall value. FPN has the best recall in Resnet-101

and consistently better than its smaller counterparts,

ResNet-18 and ResNet-50.

The results of change detection on Farm pond

dried/wetted (Task-2) taking ResNet-18, ResNet-50

and ResNet-101 as encoders are given in Table 8.

We can notice that Unet has the best F-Score of

0.9171 and the highest precision for ResNet-18. For

ResNet-50 encoder, we can observe that FPN has

the best F-Score of 0.9125, though DeepLabV3 has

best recall and UperNet has best precision values.

DeepLabV3+ has the highest recall value with com-

paratively smaller precision than the best performing

model. If we observe carefully, Task-2 has higher

overall scores irrespective of the encoder architecture.

This may be due to the presence of farm ponds at

the same locations in both temporal images where the

change is registered and also due to the fact that the

number of positives are much smaller than the num-

ber of negative images for the task.

FPCD: An Open Aerial VHR Dataset for Farm Pond Change Detection

867

The results of combining change detection tasks

(Task-1 and Task-2) as Task-3 taking ResNet-18,

ResNet-50 and ResNet-101 as encoders are given in

Table 9. For the ResNet-18 encoder, we can notice

that BiT has the best F-score of 0.8741 and the high-

est recall. For the ResNet-50 encoder, we can notice

that PSPNet has the best precision but with poor recall

values affecting the F-score.

We also conducted an ablation study on Task 3

to check if the absence of negative examples in the

training set could affect the change detection task per-

formed. We can see that BiT achieves the best F-

Score with ResNet-18 as the encoder, but we can also

see that there is a decrease in precision when com-

pared to the best model in the complete dataset, de-

spite better recall values. The results for F-scores

also show a similar trend, suggesting that even with

a high ratio, the performance is comparable even in

the absence of negative images in the train set. Sim-

ilarly, we observe that UNet++ with the best F-Score

trails the model trained on the entire dataset by only a

few points; this further supports the idea that positive

samples aid in learning while negative images may

increase the model’s robustness.

In this section, we analyze the performance of

object detection and instance segmentation task on

the Farm pond OD/IS dataset. The results of com-

parison of object detection on various existing mod-

els are given in Table 11. For the object detection

task, Probabilistic Anchor Assignment (PAA) model

achieves the best performance of 0.575 mAP with

IoU(0.5:0.95) with the ResNet-50 backbone. We

can notice that empirical attention with the atten-

tion component (0010) achieves the best mAP un-

der IoU(0.75). Guided Anchoring with FasterRCNN

model performs the best in bbox mAP(0.5) using

ResNet-50 backbone. For the instance segmenta-

tion task, many models have the capability to gener-

ate segmentation masks along with bounding boxes

results and the COCO metric for both are given

in the Table 10. We can notice that CascadeR-

CNN with ResNet-50 performs the best with 0.577

bbox mAP(0.5:0.95) and 0.694 bbox mAP(0.75) in-

dicating it to be a robust detector. Swin Trans-

former has the best bbox mAP(0.5) with its own

independent backbone. For instance segmentation,

Deformable ConvNetsv2 with MRCNN model using

ResNet-50 as backbone performs the best on segm

mAP metric with IoU(0.5:0.95), while MaskRCNN

with ResNet-50 performs the best in segm mAP(0.75)

and Swin Transformer performs the best under the

segm mAP(0.5) metric.

Table 10: Instance segmentation results on OD/IS dataset.

Method Backbone

bbox mAP

(0.5:0.95)

bbox mAP

(0.5)

bbox mAP

(0.75)

segm mAP

(0.50:0.95)

segm mAP

(0.5)

segm mAP

(0.75)

Mask RCNN

ResNet - 50 0.558 0.834 0.68 0.59 0.853 0.696

Swin Transformer Swin Transformer

0.563 0.851 0.667 0.59 0.874 0.671

Cascade RCNN

ResNet - 50 0.577 0.81 0.688 0.575 0.829 0.693

ResNet - 101 0.563 0.813 0.694 0.572 0.827 0.685

Hybrid Task Cascade

ResNet - 50 0.533 0.785 0.663 0.541 0.805 0.643

Deformable

ConvNets v2 (MRCNN)

ResNet - 50 0.571 0.833 0.673 0.591 0.859 0.691

5 CONCLUSION

The availability of high-resolution images due to ad-

vances in remote sensing capabilities has led to the

use of deep learning models for change detection. In

this paper, we introduced FPCD, a publicly available

dataset for change detection tasks for minor irriga-

tion structures - farm ponds. We also introduced a

small-scale public dataset for object detection and in-

stance segmentation of four farm pond categories. Fu-

ture works can address the issues associated with lim-

ited data availability and class imbalance issues in the

dataset.

Table 11: Object detection results on OD/IS dataset.

Method Model Name Backbone

bbox mAP

(0.5:0.95)

bbox mAP

(0.5)

bbox mAP

(0.75)

Grid RCNN Grid RCNN

ResNet - 50 0.56 0.553 0.551

Deformable DETR 2-stg Def. DETR

ResNet - 50 0.526 0.774 0.624

Faster RCNN Faster RCNN

ResNet - 50 0.555 0.832 0.646

Empirical Attention

AC-0010

ResNet - 50 0.553 0.838 0.687

AC-1111

ResNet - 50 0.557 0.829 0.663

GFL

GFL ResNet - 50 0.553 0.815 0.653

GH SSD

GHM ResNet - 50 0.528 0.787 0.629

Guided Anchoring

Faster RCNN

ResNet - 50 0.566 0.843 0.672

RetinaNet ResNet - 50 0.547 0.832 0.637

PAA

PAA ResNet - 50 0.575 0.832 0.686

YOLOv3

YOLOv3 MobileNetv2 0.437 0.729 0.485

REFERENCES

Bourdis, N., Marraud, D., and Sahbi, H. (2011). Con-

strained optical ﬂow for aerial image change detec-

tion. In 2011 IEEE IGARSS.

Chaurasia, A. and Culurciello, E. (2017). Linknet: Exploit-

ing encoder representations for efﬁcient semantic seg-

mentation. 2017 IEEE VCIP.

Chen, H. and Shi, Z. (2020). A spatial-temporal attention-

based method and a new dataset for remote sensing

image change detection. Remote Sensing, 12(10).

Chen, K., Wang, J., Pang, and et. al. (2019). MMDetec-

tion: Open MMLab Detection Toolbox and Bench-

mark. arXiv:1906.07155.

Chen, L.-C., Papandreou, G., Schroff, F., and Adam, H.

(2017). Rethinking atrous convolution for semantic

image segmentation. ArXiv, abs/1706.05587.

Chen, L.-C., Zhu, Y., Papandreou, G., Schroff, F., and

Adam, H. (2018). Encoder-decoder with atrous sep-

arable convolution for semantic image segmentation.

In Proceedings of ECCV.

Data{Meet} (2019). Indian village boundaries project.

URL: http://projects.datameet.org/indian village

boundaries/. Online; Accessed 13-June-2021.

VISAPP 2023 - 18th International Conference on Computer Vision Theory and Applications

868

Daudt, R. C., Le Saux, B., Boulch, A., and Gousseau, Y.

(2018). Urban change detection for multispectral earth

observation using convolutional neural networks. In

2018 IEEE IGARSS.

Fan, T., Wang, G., Li, Y., and Wang, H. (2020). Ma-net:

A multi-scale attention network for liver and tumor

segmentation. IEEE Access, 8:179656–179665.

Hao Chen, Z. Q. and Shi, Z. (2021). Remote sensing image

change detection with transformers. IEEE TGRS.

Ji, S., Wei, S., and Lu, M. (2019). Fully convolutional

networks for multisource building extraction from an

open aerial and satellite imagery data set. IEEE TGRS.

Kaiyu Li, Fulin Sun, X. L. (2021). Change de-

tection pytorch. URL: https://github.com/likyoo/

change detection.pytorch.

Kumar, R., Dabral, R., and Sivakumar, G. (2021).

Learning unsupervised cross-domain image-to-image

translation using a shared discriminator. CoRR,

abs/2102.04699.

Li, B., Liu, Y., and Wang, X. (2019). Gradient harmonized

single-stage detector. In AAAI Conference on Artiﬁ-

cial Intelligence.

Li, H., Xiong, P., An, J., and Wang, L. (2018). Pyramid

attention network for semantic segmentation. arXiv

preprint arXiv:1805.10180.

Li, X., Yang, J., and et. al (2020). Generalized Focal Loss:

Learning Qualiﬁed and Distributed Bounding Boxes

for Dense Object Detection. arXiv:2006.04388.

Lin, T.-Y., Maire, M., Belongie, S., Bourdev, L., Girshick,

R., Hays, J., Perona, P., Ramanan, D., Zitnick, C. L.,

and Doll

ar, P. (2014). Microsoft coco: Common ob-

jects in context.

Liu, M., Chai, Z., Deng, H., and Liu, R. (2022). A

cnn-transformer network with multi-scale context ag-

gregation for ﬁne-grained cropland change detection.

IEEE Journal of Selected Topics in Applied Earth Ob-

servations and Remote Sensing.

Lu, X., Li, B., Yue, Y., Li, Q., and Yan, J. (2019). Grid

r-cnn. In Proceedings of the IEEE Conference on

CVPR.

Maharashtra Remote Sensing Application Centre

(MRSAC) (2015). Jalyukt shivar , water con-

servation department, mantralaya, mumbai.

Source:http://mrsac.maharashtra.gov.in/jalyukt/.

Online; Accessed: 2019-09-15.

Prasad, P., Damani, O. P., and Sohoni, M. (2022). How can

resource-level thresholds guide sustainable intensiﬁ-

cation of agriculture at farm level? a system dynamics

study of farm-pond based intensiﬁcation. Agricultural

Water Management.

Redmon, J. and Farhadi, A. (2018). Yolov3: An incremental

improvement. CoRR, abs/1804.02767.

Ren, S., He, K., Girshick, R., and Sun, J. (2017). Faster

r-cnn: Towards real-time object detection with region

proposal networks. In IEEE TPAMI, number 6.

Ronneberger, O., Fischer, P., and Brox, T. (2015). U-net:

Convolutional networks for biomedical image seg-

mentation. In MICCAI.

Russell, B. C., Torralba, A., Murphy, K. P., and Freeman,

W. T. (2008). Labelme: a database and web-based

tool for image annotation. IJCV.

Shao, R., Du, C., Chen, H., and Li, J. (2021). Sunet: Change

detection for heterogeneous remote sensing images

from satellite and uav using a dual-channel fully con-

volution network. Remote Sensing.

Shen, L., Lu, Y., Chen, H., Wei, H., Xie, D., Yue, J., Chen,

R., Lv, S., and Jiang, B. (2021). S2looking: A satel-

lite side-looking dataset for building change detection.

Remote Sensing, 13(24).

Shi, Q., Liu, M., Li, S., Liu, X., Wang, F., and Zhang, L.

(2022). A deeply supervised attention metric-based

network and an open aerial image dataset for remote

sensing change detection. IEEE TGRS.

Tian, S., Zheng, Z., Ma, A., and Zhong, Y. (2020).

Hi-ucd: A large-scale dataset for urban semantic

change detection in remote sensing imagery. CoRR,

abs/2011.03247.

Tundia., C., Kumar., R., Damani., O., and Sivakumar., G.

(2022). The mis check-dam dataset for object detec-

tion and instance segmentation tasks. In VISAPP.

Tundia, C., Tank, P., and Damani, O. (2020). Aiding Irri-

gation Census in Developing Countries by Detecting

Minor Irrigation Structures from Satellite Imagery. In

GISTAM.

Van Etten, A., Hogan, D., Manso, J. M., Shermeyer, J.,

Weir, N., and Lewis, R. (2021). The multi-temporal

urban development spacenet dataset. In Proceedings

of the IEEE/CVF CVPR.

Wu, C., Zhang, L., and Du, B. (2017). Kernel slow feature

analysis for scene change detection. IEEE TGRS.

Xiao, T., Liu, Y., Zhou, B., Jiang, Y., and Sun, J. (2018).

Uniﬁed perceptual parsing for scene understanding. In

Proceedings of the ECCV.

Zhan, Y., Fu, K., Yan, M., Sun, X., Wang, H., and Qiu,

X. (2017). Change detection based on deep siamese

convolutional network for optical aerial images. IEEE

Geoscience and Remote Sensing Letters.

Zhang, C., Yue, P., Tapete, D., Jiang, L., Shangguan, B.,

Huang, L., and Liu, G. (2020). A deeply supervised

image fusion network for change detection in high res-

olution bi-temporal remote sensing images. ISPRS

Journal of Photogrammetry and Remote Sensing.

Zhang, M., Xu, G., Chen, K., Yan, M., and Sun, X. (2019).

Triplet-based semantic relation learning for aerial re-

mote sensing image change detection. IEEE Geo-

science and Remote Sensing Letters.

Zhao, H., Shi, J., Qi, X., Wang, X., and Jia, J. (2017). Pyra-

mid scene parsing network. In Proceedings of the

IEEE conference on CVPR.

Zhou, Z., Rahman Siddiquee, M. M., Tajbakhsh, N., and

Liang, J. (2018). Unet++: A nested u-net architecture

for medical image segmentation. In Deep learning in

medical image analysis and multimodal learning for

clinical decision support.

FPCD: An Open Aerial VHR Dataset for Farm Pond Change Detection

869