Pest YOLO: An Effective Insect Target Detection Algorithm for Small Targets

Xuchuan Qin

2024

Abstract

This paper proposes three distinct strategies to enhance the performance of the You Only Look Once version 5 (YOLOv5) model in object detection tasks. The enhancements encompass the integration of a BiFormer attention mechanism, the addition of an Adaptive Feature Pyramid Network (AFPN), and the replacement of the Spatial Pyramid Pooling Module (SPPF) with the Multi-Task Spatial Pyramid Pooling (MTSPPF). The BiFormer attention mechanism aims to enhance the model's focus on target regions, leading to improved detection accuracy by capturing long-range dependencies and enhancing the understanding of spatial relationships within images. Integrating AFPN into the YOLOv5 model optimizes the feature pyramid network, enabling adaptive adjustments of feature representations across various scales, which improves the detection of objects with different sizes and complexities. Additionally, the replacement of SPPF with MTSPPF facilitates more effective aggregation of spatial information from multiple scales, thereby enhancing performance while reducing both parameter count and computational complexity. Experimental evaluations on standard datasets indicate significant improvements in object detection performance for all three approaches. Collectively, these enhancements tackle challenges related to complex scenes and varying object scales, providing a comprehensive solution for improving the YOLOv5 model's effectiveness in object detection tasks.

Paper Citation

in Harvard Style

Qin X. (2024). Pest YOLO: An Effective Insect Target Detection Algorithm for Small Targets. In Proceedings of the 2nd International Conference on Data Analysis and Machine Learning - Volume 1: DAML; ISBN 978-989-758-754-2, SciTePress, pages 275-284. DOI: 10.5220/0013515700004619

in Bibtex Style

@conference{daml24,
author={Xuchuan Qin},
title={Pest YOLO: An Effective Insect Target Detection Algorithm for Small Targets},
booktitle={Proceedings of the 2nd International Conference on Data Analysis and Machine Learning - Volume 1: DAML},
year={2024},
pages={275-284},
publisher={SciTePress},
organization={INSTICC},
doi={10.5220/0013515700004619},
isbn={978-989-758-754-2},
}

in EndNote Style

TY - CONF

JO - Proceedings of the 2nd International Conference on Data Analysis and Machine Learning - Volume 1: DAML
TI - Pest YOLO: An Effective Insect Target Detection Algorithm for Small Targets
SN - 978-989-758-754-2
AU - Qin X.
PY - 2024
SP - 275
EP - 284
DO - 10.5220/0013515700004619
PB - SciTePress