Automatic butterfly identification in field environment based on YOLOV7 improved algorithm

This paper is based on the yolov7 improved algorithm for butterfly image recognition. First, we use various data augmentation algorithms to expand the data, and a self-supervised method to further expand the dataset. Moreover, we improve the original convolution module by replacing the original silu activation function, and improve the feature fusion network of yolov7 so that the network can obtain more detailed information and multi-scale information. We add the attention mechanism module to yolov7 to further improve the performance of the model. With our improved yolov7 model we are able to achieve better recognition of butterfly images in the field environment. A self-supervised approach is used in the training process to utilize the dataset that can be collected to the maximum extent, which enhances the accuracy of the model and the generalization ability of the model.