A Pyramidal Feature Fusion Model on Swimming Crab Portunus trituberculatus Re-identification

Abstract

Swimming crab Portunus trituberculatus is a vital crab species in coastal areas of China. In this study, an individual re-identification method based on Pyramidal Feature Fusion Model (PFFM) for P. trituberculatus was proposed. This method took the carapace texture of P. trituberculatus as a “biological fingerprint” and extracted carapace texture features, including global features and local features, to identify P. trituberculatus. Furthermore, this method utilized a weight adaptive module to improve re-identification (ReID) accuracy for the P. trituberculatus individuals with the incomplete carapace. To strengthen the discrimination of the extracted features, triplet loss was adopted in the model training process to improve the effectiveness of P. trituberculatus ReID. Furthermore, three experiments, i.e., PFFM on the effect of pyramidal model, P. trituberculatus features analysis, and comparisons to the State-of-the-Arts, were carried out to evaluate PFFM performance. The results showed that the mean average precision (mAP) and Rank-1 values of the proposed method reached 93.2 and 93% in the left half occlusion case, and mAP and Rank-1 values reached 71.8 and 75.4% in the upper half occlusion case. By using the experiments, the effectiveness and robustness of the proposed method were verified.