Small Target Pest Detection Algorithm Based on Channel and Spatial Collaborative Attention

• Published in: 2024 4th Asia-Pacific Conference on Communications Technology and Computer Science (ACCTCS) pp. 319 - 325
• Main Authors: Liu, Qiuyang, He, Lijun, Li, Fan
• Format: Conference Proceeding
• Language: English
• Published: IEEE 24.02.2024
• Subjects: Accuracy; Attention mechanism; Attention mechanisms; Collaboration; Crops; Network architecture; Object detection; Pest detection; Production; Small target detection; YOLOv8
• DOI: 10.1109/ACCTCS61748.2024.00063

Crop pest is one of the unavoidable problems in agricultural production, which can cause serious yield loss and quality reduction, and even jeopardize the growth and development of crops and life safety. However, conventional target detection algorithms for pest detection generally suffer from generally small target sizes, complex backgrounds, and high leakage rates. To address the issues, we propose a small target pest detection algorithm CSCA-YOLO (Channel and Spatial Collaborative Attention-YOLO) based on channel and spatial cooperative attention. Aiming at the problem of small target size for pest detection in real field environments, a small target detection head is designed. And based on the integration characteristics of small target pests, a lightweight and efficient attention mechanism named CSCA is proposed, which adopts a tri-branch design to concurrently deduce crucial information across channel, height, and width dimensions. Attention, and feature fusion with adaptive weights significantly improve the small target pest detection accuracy with low computational overhead.