Lightweight crop pest identification algorithm under natural background

Aiming at the problem of poor detection effect and low recognition accuracy of small target insects under the background of complex natural environment, this paper proposes an improved Yolo v5s insect recognition algorithm. First, the channel attention mechanism is embedded in the backbone network to improve the feature extraction ability of the algorithm for small targets. Second, the Adaptive Spatial Feature Fusion (ASFF) structure is introduced in the PANet part, and dynamic weight parameters are used to assign different weights to feature maps of different scales, so as to filter out other levels of features such as the complex environment where insects are located and improve the recognition accuracy. Finally, we change the loss function and non-maximum suppression strategy to improve the accuracy of bounding box positioning and the speed of regression. Experimental results show that the improved algorithm has a final average accuracy (mAP@0.5) of 97.8% in the D0 data set and an average detection speed of 13.66 ms per image, which is more suitable for deployment in mobile and embedded devices to achieve real-time detection.