A Systematic Approach of a Various Natural Acid‑Base Indicator on

Fruits Using CNN

Nithya C.

, Elarmathi S.

, Kavya G. S.

, Divya Bharathi G.

Kumarganesh S.

and Malathi Murugesan

Department of CSBS, Knowledge Institute of Technology, Salem 637504, Tamil Nadu, India

Department of ECE, Knowledge Institute of Technology, Salem 637504, Tamil Nadu, India

Department of BME, Paavai College of Engineering, Namakkal 637018, Tamil Nadu, India

Department of CSE, Mahendra Institute of Technology, Namakkal 637503, Tamil Nadu, India

Department of ECE, E.G.S. Pillay Engineering College, Nagapattinam 611002, Tamil Nadu, India

Keywords: Bayesian Optimization, Citrus, CNN Regression, Fluorescence, Fruits, MobileNetV2, Acid‑Base Indicators,

pH.

Abstract: To determine the ripeness of citrus fruits, one of the most widely used indicators of fruit development is the

Brix/acid ratio, which is the ratio of the fruit's sugar or soluble solids content to its acid content. A single

spectrum can be subjected to successive applications of a diverse set of statistical models known as SGFP.

These statistical models comprise particular models for differentiation of product categories, fruit type

differentiation amongst citrus varieties, separation of multiple fruit types, and characterization of

compositional differences between two sets of items that are otherwise highly similar. The Brix/acid ratio was

estimated using fluorescence spectroscopy, a technique that is not only fast and sensitive but also very

affordable. After the orange peels were removed, each peel's fluorescence value was calculated. The suggested

system is unusual in that it analyses the fluorescence spectrum using a convolutional neural network (CNN).

When performing fluorescence spectroscopy, a matrix known as an excitation and emission matrix (EEM)

can be created. For every excitation and emission wavelength, the fluorescence intensity is noted in this

matrix. A CNN was then used to perform a regression in order to determine the Brix/acid ratio of the juice

that was collected from the meat. To do this, the EEM was viewed as a picture (CNN regression).

1 INTRODUCTION

Various types of synthetic chemical indicators are

accessible for the different kinds of titrimetric

investigations. PH indicators are measures of acidity

and baseness. Acid-base indicators are chemicals

(dyes) whose colours vary when the pH of their

environment changes. They are often weak bases or

acids, which means they only dissociate and give ions

in solution to a small extent. Pure volumetric

analysis is one of the principal techniques of

quantitative methodologies. The equivalence point

calculation in titrimetry is often based on the end

point of the titration. One of the applications of this

kind of analysis is classical titrimetry, where the end

of a point is detected by adding some compounds to

the analyse solution that cause chromatic changes,

immediately after to each point. These compounds

are often called indicators. Different types of

titrimetric studies have several types of indicators,

most of which are either very weak organic acids or

very weak organic bases that respectively accept or

release electrons. Morimoto T, et al., 1994, Although

there are modern automated titration devices that

identify the equivalent point between reagaging

species, indicators are widely used for simple

titration in academic and research labs. Momin A.M.,

et al., 2010, Thus, the price of commercial indicators

is very high, and some have proved to have

deleterious effects on users as well as the risk of

polluting the environment. These properties

motivated the search for supplementary predictors

based essentially on natural forcings.

One of the most well-known and widely eaten

types of fruit all around the globe is citrus. Citrus

production was the second-highest of all fruits in

Japan in 2015, with Satsuma mandarin (Citrus unshiu

Marc.) being the most popular cultivar. Citrus must

330

C., N., S., E., S., K. G., G., D. B., S., K. and Murugesan, M.

A Systematic Approach of a Various Natural Acid-Base Indicator on Fruits Using CNN.

DOI: 10.5220/0013882600004919

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 1st International Conference on Research and Development in Information, Communication, and Computing Technologies (ICRDICCT‘25 2025) - Volume 2, pages

330-336

ISBN: 978-989-758-777-1

not only be consistently available but also of an ever-

increasingly high quality to satisfy customer demand.

To begin, the harvest period of citrus should be

carefully decided by taking into consideration the

maturity of the fruit in order to satisfy demand. Reid

M.S. 2002, This is due to the fact that when citrus fruit

is harvested with a high degree of maturity, it is more

prone to being damaged by mechanical means or

infected by pathogens during the postharvest

handling process. On the other hand, if it hasn't been

matured properly, it won't have a pleasant taste or

look, and it may not even be sellable.

Cary P.R; Iglesias D.J, 2007, Changes in size,

shape, colour, hardness, and the Brix/acid ratio have

all been used frequently to determine the maturation

level of citrus fruit. Since brix and acidity measure the

quantity of sugar or soluble solids present and the

amount of acid present, respectively, they are

especially important parameters. Kimball D, 1991,

Their ratio, sometimes referred to as the Brix/acid

ratio, is one of the most widely used indicators to

assess the ripeness of fruit in addition to the quality

of the juice.

Figure 1: Sample Images of Forty Types of Fruits.

Kondo N, et al, 2000 The fruits were planted in a

revolving motor shaft at a speed of three revolutions

per minute (rmp), and a brief video recording of

twenty seconds was recorded for each class. These

stills are taken directly from the video. Because of the

different lighting circumstances, an algorithm was

used to eliminate the backdrop from each of the

photographs. The example photos of fruit that were

taken are shown in Figure 1. The exact calculation of

the Brix/acid ratio is used to identify the optimal time

to harvest the crop. In the past, the interior quality was

evaluated based on the characteristics of the object's

exterior, such as its dimensions, form, and mass, as

well as its colour. However, the Brix and acidity

levels could not be accurately estimated. Antonucci et

al. used a portable visible-near infrared (VIS-NIR)

spectrophotometer to determine the Brix and acidity

values; however, this method is very expensive.

Tamilarasi M, et al, 2024 Another method that has

recently gained popularity is fluorescence

spectroscopy, which is a rapid, sensitive, and cost-

effective technology.

The researchers Muharfiza et al. looked into the

viability of fluorescence spectroscopy as a method for

determining the stage of maturation reached by

Satsuma mandarins. The fluorescence features were

observed and measured throughout the phases of

development and maturation, and then contrasted

with the conventional maturity indicator known as the

Brix/acid ratio. Muharfiza; et al., 2017 According to

the findings of the research, the Brix/acid ratio was

linked to the fluorescence peaks of amino acids and

chlorophyll, and the peak intensity was useful for

estimating the Brix/acid ratio. However, as of yet, no

quantitative analysis of the accuracy of the Brix/acid

ratio calculation has been conducted. In addition, the

assessment was based on just two of the fluorescence

peaks that a citrus peel really possesses, despite the

fact that a citrus peel actually contains several

fluorescence peaks. In addition, the estimate accuracy

might be improved by taking into account data other

than the peaks of the fluorescence spectrum, such as

the shoulder, unevenness, and troughs of the spectrum

Wang, et al, 2010.

2 EXPERIMENTAL SETUPS

The purpose of this study is to develop an optimum

classification model that can recognise and

distinguish between the many kinds of fruits. Python

3.0 running on Windows 10 was used to build the

suggested model, and the system setup included an i7

CPU and 16 gigabytes of random access memory

(RAM). As can be seen in Figure 2, the model is

evaluated with a variety of records abstracted from a

2 by 2 confusion matrix.

• Macro Average: This macro average is

calculated by computing F-1 for each label and

then averaging the scores, ignoring the fraction

of the dataset that corresponds to the labels.

• Weighted Average: The one that calculates F-1

of each label and averages them weighted by the

proportion of the dataset represented by each

label.

A Systematic Approach of a Various Natural Acid-Base Indicator on Fruits Using CNN

331

2.1 Estimation of the Brix/Acid Ratio

with CNN Regression

In the present study, one of the most conventional

type of CNN was used. In this particular CNN, there

are three different types of layers the fully-connected

layer, pooling layer, and the convolution layer.

Sugiyama J, et al, 2013 CNN regression was

performed by moving the regression layer to the very

last layer to get the Brix/acid estimate. 1) Mean

square error refereed as a loss function. We then

adjusted the weight of each filter and bias to

minimize mean square error due to the difference

between the estimated and actual Brix/acid ratio and

the estimated value.

Figure 2: An Illustration of a Five-Observation Gaussian

Process Regression (Dotted Line). the 95% Prediction

Interval Is Shown by the Patched Area.

In Figure 2, a dotted line represents the results of the

predictions made using a Gaussian process

regression. Barbedo J.G.A, 2018 The estimated value

of the function with the most probable outcomes is

shown in this line. The area that has been patched

with grey may contain the function's value. This

illustrates that the percentage is 95%. The Gaussian

distribution may be described using each graph that

was extracted in the vertical direction for a particular

x value.

2.2 Flowers, Plants and Fruits

Materials

Red cabbage, tulip petals, rose petals, rosa

damascene, red onion skin, curcuma, cinnamon,

ginger, saffron, black pepper, red pepper, yellow

pepper, coffee, quince leaf, strawberry, sour berry,

cornelian cherry, carrot, green walnut, parsley,

coriander, borage, and allium ampeloprasum were

prepared from Agricultural Research Center of

Tabriz. Flowers, plant leaves and petals, and the skins

of fruits and vegetables are all included as examples

Suh, et al., 2018. After being rinsed completely under

running tap water and then cleansed with distilled

water, the samples were allowed to air dry before

being pulverised using a mechanical blender.

Then, the score with respect to the p-th region by

averaging the response over the modalities:





  







(1)

Given this threshold, the precision (P), recall (R) and

F1 score are computed as:

 





(2)















(3)

 





(4)

where T

is the numeral value of true positives

(correct detections), F

is the numeral value of false

positives (false detection), F

is the number of false

negatives (miss) and T

is the numeral value of true

negatives (correct rejection).

2.3 Chemicals Required

All of the chemicals were of analytical reagent quality

and were obtained from Sigma-Aldrich Chemical

(carbon tetrachloride, chloroform, ethanol, methanol,

toluene, and hydrochloric acid) and Merck (all other

compounds). Water that had undergone two

distillations was used to make each solution.

Figure 3: Chemical Construction of Curcumin in Enol (A)

and Keto (B).

2.4 Apparatus

Reagent bottle, weighing balance, spatulas, hot plate,

shaker, oven, electric blender, test tubes, test tube

support, droppers, 50 mL buretes, wash bottle,

beaker, spatula, pipettes, pipette filler, funnel, clamp

support, tissue, magnetic stirrer, watch glass,

volumetric cylinders of 25 mL and 50 mL, conical

ICRDICCT‘25 2025 - INTERNATIONAL CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION,

COMMUNICATION, AND COMPUTING TECHNOLOGIES

332

flask, pH paper, magnetic stirrer, watch glass pH

meter (electronic - Switzerland model OHAUS2100),

measuring cylinders, and glass and calomel

electrodes.

2.5 Extraction Preparation of Flowers,

Plants and Fruits in Various

Solvents

To explore the extraction efficiencies of the five

organic solvents, two grams of the sample powder

were mixed with fifty millilitres of carbon

tetrachloride, chloroform, ethanol, methanol, and

toluene as solvents by stirring the mixture for forty-

eight hours. The solution was vigorously agitated and

stirred to ensure the entire component would

dissolve. The extract is filtered by suction through

strainer paper with the aid of a Buchner funnel and

collected in a filter flask. All aqueous abstract were

used after evaporation as natural pointer in

acidimetric and alkalimetric measurements. One-fifth

of the volume of the original abstract was extracted.

The extract was protected from light by storing it in a

container with a lid in the dark.

3 RESULTS AND DISCUSSION

Figure 4: Relationship Between the Number of

Optimizations and the Brix/Acid Ratio Estimate's Smallest

Absolute Inaccuracy.

Figure 4 illustrates the relationship between the

minimal absolute error in the Brix/acid ratio and the

number of parameter modifications. As more

improvements were carried out, the estimate's lowest

absolute error decreased to a smaller value. Results

from the same optimization method were quite

similar when it was repeated.

Figure 5: Training Accuracy and Loss in Terms of Training

and Validation of Tl-Mobilenetv2 Model.

Figure 5 also includes a representation of the

suggested model's training loss. The illustration

shows that the training loss is relatively significant at

the beginning of the training phase since the model

has not yet been presented with the data. This is

illustrated in the image. But, with time the model

learns to understand the pictures and begins to recall

them; as a result, the training loss progressively

decreases. It is possible to see that the training loss

approaches 0.6 during the first 20 iterations and then

decreases noticeably with each subsequent iteration

beyond that point. The training loss has reached 0.3

by the time the 100th iteration has passed, which

indicates that the model has the qualities of a good

one.

Table 1: Assessment Value of Brix/Acid Ratio.

S. No

Assessment

Method

AEE of

Brix/Acid

Ratio

Two peaks of poly

methoxy flavone

6.20

Two peaks of

Tryptophan and

Chlorophyll

4.48

PCR (Principal

Component

Regression)

4.03

Proposed method

(CNN regression)

2.47

Together with the absolute error of the present CNN

regression and the previous estimate methods, Table

1 displays the absolute error of the Brix/acid ratio.

A Systematic Approach of a Various Natural Acid-Base Indicator on Fruits Using CNN

333

Figure 6: The Phenotypic Alteration of Post-Harvest Tomato Fruits Treated With C2h4 or C2h4-H2s.

(A) Morphological change of tomato fruit after

harvest. (B) The tomato post-harvest storage a/b

value regarding colour parameter changes after C2H4

or C2H4-H2S treatment. The magenta to green colour

range is represented by the a* value whereas the

yellow to blue colour range is represented by the b*

value. Values are the means standard deviations (n

= 3). All of the above experiments were performed at

room temperature and 85–90% relative humidity.

The symbols and * indicate a p 0.05 and p 0.01,

respectively, significant difference between C2H4

and C2H4-H2S.

Figure 6A depicts the colour change of tomato

fruit over the post-harvest storage period. On the 0th

day, all tomato fruits are clearly seen to be at the

green-yellow (G-Y, 30% yellow skin) stage. During

the first and second days, some C2H4-induced

tomatoes are in the green-orange (G-O, 50% orange

skin) stage, but tomato fruits are practically in the G-

Y stage due to the C2H4 + H2S co-treatment. The

colour change was indicated using an a/b value, as

seen in Figure 6B. Over the storage time, the a/b

values of the two treatment groups increased, whereas

the a/b value of the H2S + ethylene treatment group

remained lower than that of the ethylene treatment

alone. Hence, further H2S treatment may reduce

tomato colour change and postpone tomato fruit

ripening during post-harvest storage.

Table 2: The Factors Score of All the Metabolites by Principal Component Analysis in Tomato Fruits.

Metabolites

Component 1

Component 2

Component 3

Content of titratable acid

(C2H4 + H2S)

0.975

0.18

0.150

Protease activity (C2H4)

-0.95

0.12

0.272

Content of chlorophyll b

(C2H4 + H2S)

0.948

0.101

-0.25

Content of starch (C2H4

+ H2S)

0.939

0.156

-0.228

Content of titratable acid

(C2H4)

0.895

0.427

0.240

Content of Anthocyanidin

(C2H4 + H2S)

-0.88

0.191

0.276

Protease activity (C2H4

+ H2S)

-0.871

0.276

0.376

Content of total phenol

(C2H4 + H2S)

-0.806

0.535

0.133

Content of chlorophyll a

(C2H4)

0.796

0.522

-0.109

Content of reducing

sugar (C2H4)

0.783

0.420

-0.537

ICRDICCT‘25 2025 - INTERNATIONAL CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION,

COMMUNICATION, AND COMPUTING TECHNOLOGIES

334

Content of reducing

sugar

(C2H4 + H2S)

0.776

0.252

-0.529

Content of total phenol

(C2H4)

-0.697

0.427

0.237

Content of chlorophyll

(C2H4)

0.69

0.645

0.267

Content of Anthocyanidin

(C2H4)

-0.633

0.581

0.445

Content of flavonoid

(C2H4 + H2S)

0.221

0.952

-0.274

This is a principal component analysis study

presented in Table 2 and seen in Figure 7. 77% of the

variance was provided by the first three components

(PC1, PC2, and PC3 with respective contribution

rates being 47.64, 38.42, and 6.3%. The major

components described by PC1 were TA, protease

activity and starch. The major contributor to PC2 was

Flavonoid, chlorophyll b, and amylase activity, while

ascorbic acid accounted for PC3.

Table 3: Anthocyanidic Compounds, Obtained from the Analysis in

the Uhplc-Esi-Q-Orbitrap-Ms / Ms.

Compound

Name

Fragment

Weight

(m/z)

Condensed

Formula

Retention

time

(Min)

Cyanidin -

Diglucoside

611.18

m/z

19.1 min

Cyanidin –

Monoglycoside

433.11

m/z

20.2 min

Pelargonidin –

Diglucoside

595.18

m/z

19.5 min

Cyanidin

287.06

m/z

19.1 min

Pelargonidin

271.06

m/z

19.5 min

Table 4: Values of Pk1 As a Function of Ph Stability in the Presence

of Oxygen and Light from the Extract Punica Granatum L.

Number

Absorbance

Pk1

0.455

a = Ain

6.5

0.461

b=Hin-

0.473

7.5

0.55

0.578

1.5

1.018

a = Ain

0.305

b=Hin-

The characterization of the extract of the fruit Punica

granatum L. was carried out using a UV-Vis

spectrophotometer, and the investigation of the

extract's stability in the presence of light lasted for a

period of seven days (Spectro quant Pharo 300). The

different performance analyses of a UHPLC system

are shown in Figures 9 and 10. And the PH scale is

shown in figure 11 below.

Table 5: Stability in the Presence of Light As a Function of the

Absorbance of the Two Glass Bottles (Colorless and Amber).

Colorless glass

bottle

Amber glass

bottle

Days

Dates

Absorbance

15-May-2017

0.77

16-May-2017

0.593

0.366

17-May-2017

0.495

0.439

18-May-2017

0.496

0.452

19-May-2017

0.565

0.452

20-May-2017

0.533

0.416

23-May-2017

0.497

0.371

4 CONCLUSIONS

A fluorescence measurement of extracted citrus peel

was carried out, and a regression using CNN was

executed in order to provide an accurate estimate of

the Brix/acid ratio of juice taken from the flesh. It was

decided to treat the EEM that was generated from the

fluorescence measurement as a picture, which made

the CNN regression possible. As a consequence of

this, the absolute error in the Brix/acid ratio was

assessed to be 2.48, which is a significant

improvement compared to the values produced by the

various other approaches in the earlier investigations.

Not only was this a suitable strategy for the

prediction, but we also did Bayesian optimization in

order to choose hyper-parameters in the deep neural

network. Both of these things contributed to the

A Systematic Approach of a Various Natural Acid-Base Indicator on Fruits Using CNN

335

accuracy of the forecast. Because of the optimization,

the parameters were able to be determined

automatically and accurately. This was made possible

by the optimization. In addition, it wasdiscovered that

the optimization process itself was responsible for the

high level of accuracy. In future work, a mobile-based

application will be further enhanced using a larger

number of different fruits, which aims to lead to a

wider range of fruit classification.

REFERENCES

Antonucci F, Pallottino F, Paglia G, Palma A, D’Aquino S,

Menesatti P, Non-destructive estimation of mandarin

maturity status through portable VIS-NIR

spectrophotometer. Food Bioprocess. Technol. 2011, 4,

809–813.

Barbedo J.G.A, Factors influencing the use of deep learning

for plant disease recognition. Biosyst. Eng. 2018, 172,

84–91.

Barbedo J.G.A, Impact of dataset size and variety on the

effectiveness of deep learning and transfer learning for

plant disease classification. Comput. Electron. Agric.

2018, 153, 46–53.

Cary P.R, Citrus Fruit Maturity; MPKV: Rahuri, India,

1974; p. 26.

Christensen J, Povlsen, V.T, Sorensen J, Application of

fluorescence spectroscopy and chemometrics in the

evaluation of processed cheese during storage. J. Dairy

Sci. 2003, 86, 1101–1107.

Deng, L.Hinton, G, Kingsbury, B. New types of deep neural

network learning for speech recognition and related

applications: An overview. In Proceedings of the 2013

IEEE International Conference on Acoustics, Speech

and Signal Processing (ICASSP), Vancouver, BC,

Canada, 26–31 May 2013; pp. 8599–8603

Iglesias D.J, Cercós M, Colmenero-Flores J.M Naranjo,

M.A, Ríos, G Carrera E, Ruiz-Rivero O, Lliso I,

Morillon R, Tadeo F.R, et al. Physiology of citrus

fruiting. Braz. J. Plant Physiol. 2007, 19, 333–362.

Itakura K, Hosoi F, Estimation of tree structural parameters

from video frames with removal of blurred images

using machine learning. J. Agric. Meteorol. 2018, 74,

154–161.

Kimball D, Citrus Processing: Quality Control and

Technology; Springer Science & Business Media: New

York, NY, USA, 1991; p. 55.

Kondo N, Ahmad U, Monta M, Murase H, Machine vision

based quality evaluation of Iyokan orange fruit using

neural networks. Comput. Electron. Agric. 2000, 29,

135–147.

Maggiori E, Tarabalka Y, Charpiat G, Alliez P,

Convolutional neural networks for large-scale remote-

sensing image classification. IEEE Trans. Geosci.

Remote Sens. 2017, 55, 645–657.

Miao S, Wang, Z.J, Liao, R. A CNN regression approach

for real-time 2D/3D registration. IEEE Trans. Med.

Imaging 2016, 35, 1352–1363. [CrossRef]

Ministry of Agriculture, Forestry and Fisheries. The

Situation Surrounding Fruits. Available online: http:

//www.maff.go.jp/j/seisan/ryutu/fruits/attach/pdf/index

-57.pdf (accessed on 20 November 2018).

Momin A.M, Kondo N, Ogawa Y, Shiigi T, Kurita M,

Ninomiya K, Machine vision system for detecting

fluorescent area of citrus using fluorescence image.

Proc. IFAC 2010, 43, 241–244.

Morimoto T, Chikaizumi S, Hashimoto Y, Intelligent

quality control of fruit storage factory. J. Shita 1994, 6,

191–196.

Muharfiza; Riza A.F.D, Saito Y, Itakura K, Kohno Y,

Suzuki T, Kuramoto M, Kondo N, Monitoring of

Fluorescence Characteristics of Satsuma Mandarin

(Citrus unshiu Marc.) during the Maturation Period.

Horticulturae 2017, 3, 51.

P. Elayaraja, Kumarganesh S, K. Martin Sagayam & J.

Andrew, (2024), “An automated cervical cancer

diagnosis using genetic algorithm and CANFIS

approaches” International Journal of Technology and

Health Care, 32 (4), pp. 1-17, https://content.iospres

s.com /articles/technology-and-health-care/thc230926.

Reid M.S., Maturation and maturity indices. In Postharvest

Technology of Horticultural Crops; University of

California Division of Agriculture and Natural

Resources Publication: Oakland, CA, USA, 2002; pp.

21–28.

Selvalakshmi B, Hemalatha K, Kumarganesh S &

Vijayalakshmi P (2025), Performance analysis of image

retrieval system using deep learning techniques.

Network: Computation in Neural Systems, 1–21.

https://doi.org/10.1080/0954898X.2025.2451388.

Sugiyama J, Tsuta M, Discrimination and quantification

thechnology for food using fluorescence fingerprint.

Nippon Shokuhin Kagaku Kogaku Kaishi 2013, 60,

457–465.

Suh, H.K, Ijsselmuiden J, Hofstee J.W, van Henten E.J,

Transfer learning for the classification of sugar beet and

volunteer potato under field conditions. Biosyst. Eng.

2018, 174, 50–65.

Tamilarasi M, Kumarganesh S, K. Martin Sagayam and

Andrew J, (2024) “Detection and Segmentation of

Glioma Tumors Utilizing a UNet Convolutional Neural

Network Approach with Non-Subsampled Shearlet

Transform” Journal of Computational

Biology, 31 (8) pp. 1-16, https://www.liebertpub.com

/doi/full/10.1089/cmb.2023.0339.

Wang, X, Cao, L, Yang S.T, Lu F, Meziani M.J, Tian L,

Sun K.W, Bloodgood, M.A.; Sun, Y.P. Bandgap-Like

strong fluorescence in functionalized carbon

nanoparticles. Angew. Chem. Int. Ed. 2010, 49, 5310–

5314.

ICRDICCT‘25 2025 - INTERNATIONAL CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION,

COMMUNICATION, AND COMPUTING TECHNOLOGIES

336