Phenotypic parameter measurement is one of the most important methods for monitoring the growth of Larimichthys crocea and estimating its economic benefits. However, most current studies on phenotypic parameter measurements focus only on the length of the fish body, with few studies focusing on essential details such as the measurement of parameters at the caudal peduncle. Also, manual measurement is time-consuming and laborious. This study designed and developed an L. crocea phenotypic parameter measurement system using computer vision technology to measure the phenotypic parameters of L. crocea. First, an improved segmentation model, Mask-RCNN, was constructed to remove the fin from L. crocea. The improvement in the model includes two aspects: detection speed and accuracy. The model replaces the backbone network of the original network model with the MobileNetV2 network to improve detection speed. Subsequently, a hybrid attention module, CBAM, is introduced into the backbone network, and PointRend is imported into the head network. These improvements have enhanced the extraction accuracy of fish body edges. The accuracy of the optimal model was 87.94%, mAP was 83.21%, and average single-image detection time was 70.5 ms. We developed a L. crocea parameter measurement system for the greater amberjack based on PyQt5, which realized the fusion of fin removal images and complete contour images with the location of the key points and calculated the phenotypic parameters through the positional information. The average relative error for systematic measurements of the six parameters was 4.04%. The average absolute percentage error of each parameter for different sizes of L. crocea was under 7%. Overall, the process designed in this study provides a new method for measuring the phenotypic parameters of L. crocea and a new way of thinking to improve the efficiency of measuring these parameters.