YOLOv8 forestry pest recognition based on improved re-parametric convolution
Open Access
- 11 March 2025
- journal article
- research article
- Published by Frontiers Media SA in Frontiers in Plant Science
- Vol. 16, 1552853
Abstract
Introduction: The ecological and economic impacts of forest pests have intensified, particularly in remote areas. Traditional pest detection methods are often inefficient and inaccurate in complex environments, posing significant challenges for effective pest management. Enhancing the efficiency and accuracy of pest detection under resource-limited conditions has thus become a critical issue. This study aims to address these challenges by proposing an improved lightweight forestry pest detection algorithm, RSD-YOLOv8, based on YOLOv8.Methods: To improve the performance of pest detection, we introduced several modifications to the YOLOv8 architecture. First, we proposed RepLightConv to replace conventional convolution in HGNetV2, forming the Rep-HGNetV2 backbone, which significantly reduces the number of model parameters. Additionally, the neck of the model was enhanced by integrating a slim-neck structure and adding a Dyhead module before the output layer. Further optimization was achieved through model pruning, which contributed to additional lightweighting of the model. These improvements were designed to balance detection accuracy with computational efficiency, particularly for deployment in resource-constrained environments.Results: The experimental results demonstrate the effectiveness of the proposed RSD-YOLOv8 model. The model achieved a Map@0.5:0.95(%) of 88.6%, representing a 4.2% improvement over the original YOLOv8 model. Furthermore, the number of parameters was reduced by approximately 36%, the number of operations decreased by 36%, and the model size was reduced by 33%. These improvements indicate that the RSD-YOLOv8 model not only enhances detection accuracy but also significantly reduces computational burden and resource consumption.Discussion: The lightweight technology and architectural improvements introduced in this study have proven effective in enhancing pest detection accuracy while minimizing resource requirements. The RSD-YOLOv8 model's ability to operate efficiently in remote areas with limited resources makes it highly practical for real-world applications. This advancement holds positive implications for agroforestry ecology and supports the broader goals of intelligent and sustainable development. Future work could explore further optimization techniques and the application of this model to other domains requiring lightweight and accurate detection systems.Keywords
Funding Information
- Department of Science and Technology of Jilin Province (20220202036NC)
This publication has 51 references indexed in Scilit:
- Excellent tomato detector based on pruning and distillation to balance accuracy and lightweightComputers and Electronics in Agriculture, 2024
- Slim-neck by GSConv: a lightweight-design for real-time detector architecturesJournal of Real-Time Image Processing, 2024
- Infrared Object Detection Method Based on DBD-YOLOv8IEEE Access, 2023
- Lightweight tomato real-time detection method based on improved YOLO and mobile deploymentComputers and Electronics in Agriculture, 2023
- Plant diseases and pests detection based on deep learning: a reviewPlant Methods, 2021
- IPBES, 2019. Summary for policymakers of the global assessment report on biodiversity and ecosystem services of the Intergovernmental Science‐Policy Platform on Biodiversity and Ecosystem ServicesPopulation and Development Review, 2019
- Detection of stored-grain insects using deep learningComputers and Electronics in Agriculture, 2018
- Grad-CAM: Visual Explanations from Deep Networks via Gradient-Based Localization2017
- Faster R-CNN: Towards Real-Time Object Detection with Region Proposal NetworksIEEE Transactions on Pattern Analysis and Machine Intelligence, 2016
- Recent advances in biological control of important native and invasive forest pests in ChinaBiological Control, 2014