This chapter aims to streamline the analysis of long-duration dolphin videos while reducing the workload of dolphin caretakers. To address this challenge, we utilized the capabilities of DeepLabCut, an open-source toolbox designed for pose estimation, and extended its capabilities through advanced animal assembly and tracking methods, which are particularly crucial for scenarios involving multiple dolphins. It is worth noting that there is limited research focusing on behavioral analysis of dolphins at present. Dolphin behavior holds significant ecological and academic implications; however, traditional observation techniques are time-consuming and often imprecise. To tackle this issue, we proposed an innovative approach that combines DeepLabCut with ChatGPT for rapid analysis. We annotated key skeletal points on dolphins and subsequently trained a model using a pretrained ResNet neural network. Postprocessing the model’s output for body parts allows us to swiftly obtain data on dolphin behaviors. Finally, this behavior data serves as ChatGPT’s database, enabling long-term analysis of dolphin behavior patterns. Our method not only ensures precise marine behavior analysis but also contributes substantially to the conservation and research of dolphins. Experimental results demonstrate that this system achieves an impressive accuracy rate of 92.5% and can effectively detect both the skeleton and behavior of dolphins.