A Lightweight Crop Pest Detection Method Based on Convolutional Neural Networks

Existing object detection methods with many parameters and computations are not suitable for deployment on devices poor performance in agricultural environments. Therefore, this study proposes a lightweight crop pest method based convolutional neural networks, named YOLOLite-CSG. The basic architecture of the is derived from simplified version YOLOv3, namely YOLOLite, k-means++ utilized to improve generation process prior boxes. In addition, sandglass block coordinate attention used optimize structure residual blocks. was evaluated CP15 dataset. Its precision exceeds that at 82.9%, while number 5 million, only 8.1% by 9.8 GFLOPs, 15% YOLOv3. Furthermore, superior all other commonly study, maximum improvement 10.6%, it still has significant edge computation required. excellent extremely few computations. It well-suited be deployed equipment detecting pests