China's offshore squid fishing technology is relatively simple, while intelligent fishing technology is very important for offshore fishery resources statistics. In the process of intelligent squid fishing, when YOLOV3 algorithm is used for detection, the feature extraction network Darknet53 model has a large amount of parameters, and its output weight storage capacity is high, which reduce the detection speed. Therefore, in this study a lightweight MobileNetV3 network was established as a feature extraction network for preliminary effective feature extraction, and then CSP bottleneck layer was designed as an inverse bottleneck structure to improve the feature extraction ability. Finally, the loss function of network model was established by CIoU model. In the laboratory environment, squid data set was used to verify the effectiveness of lightweight network model, and the performance of its backbone network and loss function were analyzed. Through training the target data set, the lightweight network model and YOLOV3 network model structure were used for ablation experiments to verify the accuracy and practicability of lightweight network application in ocean fishing technology. The results show that the performance of lightweight network structure is obviously better than that of YOLOV3 network model structure; it can greatly reduce the number of parameters, improve the detection speed, shorten the detection time, improve the detection rate and improve the efficiency of squid real-time detection. This study provides an important reference for the investigation of pelagic fishery resources.