Shihao Guo
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Research Interests
Shihao Guo has a total of 4 co-authors. Here are the top co-authors by number of shared publications:
- Chunqi Gao
College of Information Engineering, Dalian Ocean University, Dalian 116023, China
1 shared publication - Hong Yu
College of Information Engineering, Dalian Ocean University, Dalian 116023, China
1 shared publication - Jianhao Yin
College of Information Engineering, Dalian Ocean University, Dalian 116023, China
1 shared publication - Junfeng Wu
College of Information Engineering, Dalian Ocean University, Dalian, China
1 shared publication
Latest Publications
- Open Access
- 25 October 2022
- Published by MDPI AG in Electronics
- Vol. 11 (21) , 3459
FIRN: A Novel Fish Individual Recognition Method with Accurate Detection and Attention Mechanism
Fish individual recognition technology is one of the key technologies to realize automated farming. Aiming at the deficiencies in the existing animal individual recognition technology, this paper proposes a method for individual recognition of underwater fish based on deep learning technology, which is divided into two parts: fish individual object detection and fish individual recognition. In the object detection part, the research has improved a new object detection for underwater fish based on the YOLOv4 algorithm, which changed the feature extraction network in YOLOv4 from CSP Darknet53 to Mobilenetv3 and changed the 3 × 3 convolution in the enhanced feature extraction network PANet to depthwise separable convolution. Compared with the original YOLOv4, the mean average precision is improved by 1.97%. For individual recognition, an algorithm called FIRN (Fish Individual Recognition Network) for individual recognition of underwater fish is proposed. The feature extraction network of the algorithm uses the improved ResNext50, and the loss function uses Arcface Loss. The CBAM attention module is introduced in the residual block of ResNext50, the max-pooling layer in the trunk is removed, and dilated convolution is introduced in the residual block, which increases the receptive field and improves the ability of feature extraction. Experiments show that the FIEN algorithm can enhance the compactness within a class while ensuring the separability between classes, and has a better recognition effect than other algorithms.
- 25 October 2022
- Published by MDPI AG in Electronics
- Vol. 11 (21) , 3459
Open AccessFIRN: A Novel Fish Individual Recognition Method with Accurate Detection and Attention Mechanism
Fish individual recognition technology is one of the key technologies to realize automated farming. Aiming at the deficiencies in the existing animal individual recognition technology, this paper proposes a method for individual recognition of underwater fish based on deep learning technology, which is divided into two parts: fish individual object detection and fish individual recognition. In the object detection part, the research has improved a new object detection for underwater fish based on the YOLOv4 algorithm, which changed the feature extraction network in YOLOv4 from CSP Darknet53 to Mobilenetv3 and changed the 3 × 3 convolution in the enhanced feature extraction network PANet to depthwise separable convolution. Compared with the original YOLOv4, the mean average precision is improved by 1.97%. For individual recognition, an algorithm called FIRN (Fish Individual Recognition Network) for individual recognition of underwater fish is proposed. The feature extraction network of the algorithm uses the improved ResNext50, and the loss function uses Arcface Loss. The CBAM attention module is introduced in the residual block of ResNext50, the max-pooling layer in the trunk is removed, and dilated convolution is introduced in the residual block, which increases the receptive field and improves the ability of feature extraction. Experiments show that the FIEN algorithm can enhance the compactness within a class while ensuring the separability between classes, and has a better recognition effect than other algorithms.