Ocean target recognition model based on attention mechanism and Fast R-CNN deep learning
WEN Lili, SUN Miao, WU Man
1.Information Industry Office, Guangxi Zhuang Autonomous Region Botanical Garden of Medicinal Plants, Nanning 530023, China; 2.Technology Innovation Center of Marine Information, Ministry of Natural Resources, Tianjin 300171, China; 3.National Marine Data Information Center, Tianjin 300171, China; 4.Information Department, Guangxi Zhuang Autonomous Region Academy of Oceanography, Nanning 530022, China; 5.School of Electrical Engineering, Guangxi University, Nanning 530004, China; 6.Beibu Gulf Artificial Intelligence Technology Research Institute, Nanning Normal University, Nanning 530001, China
Abstract: In order to improve the detection ability of Faster R-CNN model for complex ocean targets, an improved Faster R-CNN model for multi-scale complex ocean targets is proposed by application of attention mechanism of adaptive scale. The model combines convolution network with SKNet network to enhance the feature extraction ability and effectiveness of the model. A total of 12 000 sample databases were established to detect four typical marine targets including ship, oyster raft, mangrove and coastline and recognize by using 91 satellite image assistant and UAV high-definition image. The comparison of improved Faster R-CNN model with the original model showed that feature extraction ability and target detection ability were significantly better than the original model, with the overall recognition accuracy of 87.1%, and with the maximal ship recognition accuracy (94.4%) among the four types of typical marine targets, while the mangrove recognition accuracy was found to be 75.1% because of its complex features and unclear boundary, although the improved model slightly increases the amount of calculation. The finding shows that the Faster R-CNN model with SKNet not only enhances the feature extraction ability of multi-scale complex targets, but also is more suitable for the detection and recognition of complex marine targets.
文莉莉, 孙苗, 邬满. 基于注意力机制和Faster R-CNN深度学习的海洋目标识别模型[J]. 大连海洋大学学报, 2021, 36(5): 859-865.
WEN Lili, SUN Miao, WU Man. Ocean target recognition model based on attention mechanism and Fast R-CNN deep learning. Journal of Dalian Ocean University, 2021, 36(5): 859-865.
2021, Vol. 36 
