Insect detection from imagery using YOLOv3-based adaptive feature fusion convolution network

Context Insects are a major threat to crop production. They can infect, damage, and reduce agricultural yields. Accurate fast detection of insects will help insect control. From computer algorithm point view, from imagery is tiny object problem. Handling objects in large datasets challenging due small resolution the an image, other nuisances such as occlusion, noise, lack features. Aims Our aim was achieve high-performance detector using enhanced artificial intelligence machine learning technique. Methods We used YOLOv3 network-based framework, which high performing computationally detector. further improved original feature pyramidal network by integrating adaptive fusion module. For training network, we first applied data augmentation techniques regularise dataset. Then, trained features optimised hyper-parameters. Finally, tested proposed on subset dataset multi-class pest Pest24, contains 25 878 images. Key results achieved accuracy 72.10%, superior existing techniques, while achieving rate 63.8 images per second. Conclusions compared with several models regarding processing speed. The method performance both terms computational Implications demonstrates that networks provide foundation for developing real-time systems better control damage.