A laser projection system for wireless light transmission is designed and an modified YOLOv3 (you only look once, v3) network based on deep learning is proposed to detect the location of mouse images. The network first uses packet convolution to compress network parameters to increase target detection speed, and then uses channel shuffle to enhance the network's information flow capabilities. The ratio between the positive sample and the negative sample is adjusted by two hyperparameters on the cross entropy loss function to reduce the weight of the easily classified sample in the loss function, and the detection accuracy is improved. The experimental results on the PASCAL VOC2007 and the self-made mouse image datasets show that the proposed detection algorithm based on the improved YOLOv3 network has a detection accuracy of 90.3%, which is superior to the traditional network structure in terms of detection speed and detection accuracy. The laser projection system using the algorithm can detect moving mouse targets in real time and perform optogenetic experiments such as wireless light transmission.