A swimming crab portunus trituberculatus re-identification method based on RNN encoding of striped key regions. (April 2023)

• Record Type: Journal Article
• Authors: Zhang, Kejie; Xin, Yu; Xie, Zhijun; Shi, Ce
• Is Part Of: Engineering applications of artificial intelligence. Volume 120(2023)
• Publication Date: 2023-04
• Subjects: Re-identification -- RNN -- Keypoints detection -- Hourglass; Engineering -- Data processing -- Periodicals; Artificial intelligence -- Periodicals; Expert systems (Computer science) -- Periodicals; Ingénierie -- Informatique -- Périodiques; Intelligence artificielle -- Périodiques; Systèmes experts (Informatique) -- Périodiques; Artificial intelligence; Engineering -- Data processing; Expert systems (Computer science); Periodicals; 620.00285
• DOI: 10.1016/j.engappai.2023.105900
• Languages: English
• ISSNs: 0952-1976

Abstract

Abstract: The traditional tracing system of swimming Portunus trituberculatus uses invasive tags to identify individuals. However, the invasive tags are easily lost and vulnerable. Fortunately, with the development of intelligent imaging devices, these problems never affect the approaches of image-based re-identification (re-ID). Since the white spots on the P. trituberculatus carapace are controlled by genes, the spots can be regarded as the fingerprint of P. trituberculatus . On P. trituberculatus carapace, there are white spots sequentially distributed in some narrow-striped regions. In this paper, in order to identify individuals, we locate the striped key regions (SKRs) and novelly adopt RNN extracting features from SKRs. The process of P. trituberculatus re-ID consists of keypoints detection, SKRs segmentation, SKRs alignment, and SKRs encoding. In terms of keypoints detection and SKRs segmentation, we use Stacked Hourglass to detect keypoints on the P. trituberculatus carapace image. The detected keypoints are utilized to locate and segment 4 SKRs on P. trituberculatus carapace images, and after SKRs alignment being processed, the segmented SKRs are aligned and standardized. As standardized SKRs are narrow stripes, we use RNN to extract SKRs features for representing the SKRs. That is because the features obtained by RNN is verified more practical than CNN. Finally, experimental results show that the method we proposed outperforms other state-of-the-art re-ID algorithms on P. trituberculatus dataset.