Abstract: The Chiloscyllium plagiosum has high economic and medical value. However, the real artificial breeding conditions cannot meet the high requirements for the breeding environment of marine fish, such as water quality and temperature, often leading to large-scale illness even death. Since video imaging has been widely used to quantitatively analyze the movement behavior of farmed fish, the technique can contribute to identifying abnormal behavior for the early warning, and thereby effectively improving the level of breeding and conservation. In this study, an imaging algorithm was proposed for the semantic part segmentation of Chiloscyllium plagiosum using encoder-decoder architecture, thereby analyzing the body movement and posture of the Chiloscyllium plagiosum. Three steps were as follows: 1) The images of Chiloscyllium plagiosum were divided into 7 visible body components, according to the morphological characteristics, including the head, left pectoral fin, right pectoral fin, left ventral fin, right ventral fin, trunk, and tail. Then, the sub-images of Chiloscyllium plagiosum were extracted from the video images in the panoramic breeding surveillance under a breeding circumstance, where a total of 476 candidate patterns were obtained, while all the images in the dataset were manually marked. After that, data augmentation was used to increase the number of images, and thus a total of 1 944 images were obtained, of which 1 166 images were selected as training images, and 778 images were selected as test images. 2) The pre-processed training dataset was fed into the network model of semantic segmentation by fine-tuning network parameters, where a deep learning framework was used to optimize the network training for the best. Then, the test dataset was put into the trained model for the segmentation. 3) Post-processing was performed to fill the holes within objects or remove small objects, where a disk structure of mathematical morphology was used to calculate the areas of connected regions. Simple and effective post-processing was utilized to obtain the optimal segmentation of fish body images under complex backgrounds or interference environments. Then, the semantic part segmentations in different colors were used to locate the centroid of the fish head and trunk for the body coordinates. The posture of the target was analyzed to calculate in a single frame image, and thereby identify the movement changes of the fish body in the frame sequence. The main steps of this work included: 1) To draw the body coordinates; 2) to analyze and calculate the direction of the fish body; 3) to identify the direction of movement. Compared with the Segnet and FCN-8s network architecture for semantic part segmentation, the test dataset showed that the segmentation using the Segnet network improved the accuracy of FCN-18s network by 1.5, 4.7, 6.95, 6.56, 6.01, 0.85, and 0.84 percentage points, respectively. Semantic part segmentation can be used to effectively distinguish the action posture of Chiloscyllium plagiosum body. The finding can lay a foundation for the recognition of abnormal fish behavior and further development of animal behavior experiments for the Chiloscyllium plagiosum.