The invention belongs to the technical field of crossing of water surface unmanned platform environment sensing and control systems, in particular to a deep learning detection method based on USV water surface optical target tracking. The USV loading camera collects videos in real time and transmits video signals to the USV internal embedded computer through the image collection card, the computerselects a key frame every other n frames, then defuzzification processing is carried out on the key frame, and finally a convolutional neural network is used for detecting a water surface target. According to the invention, in a water surface target detection and positioning process; an image deblurring technology based on a convolutional neural network is added; the target fuzzy problem caused by seawater fluctuation, target object motion and USV navigation is improved, and the target detection based on the regression mode of 24 convolution layers and 2 full connection layers has the advantages of high speed and low background false detection rate, so that USV target detection has real-time performance and is not easily interfered by sea surface sunlight refraction and the like.