Marine Life Detection Algorithm Based on Deep Convolutional Generative Adversarial Network and ResNeXt50 with FPN Faster R-CNN Network

Abstract

The development and utilization of marine resources has always been a focus of attention, and a prerequisite for the rational development of marine resources is the understanding of the distribution of biological resources in the ocean, so the detection of marine life has become an inevitable challenge in the development of marine resources. The acquisition of marine life images by underwater cameras is particularly difficult due to the complexity of the underwater terrain and the difficulty of underwater imaging. Therefore, the detection of marine life often faces problems of difficulties in dataset acquisition, scarcity of data in the dataset and low quality of images. To address the above problems, a marine life detection algorithm based on the combination of deep convolutional generative adversarial network and ResNeXt50 with FPN Faster R-CNN network is proposed in this paper. The algorithm uses Faster R-CNN as the basic framework for marine life detection and uses deep convolutional generative adversarial network to generate data for small dataset to effectively expand the dataset without the difficulty of underwater image acquisition. For the feature extraction network, we use ResNeXt50, which employs the idea of group convolution, as the basic network for feature extraction. Then, the network is combined with feature pyramid network to enhance the feature extraction capability. The expanded dataset fed into the ResNeXt50 with FPN Faster R-CNN target detection network for training. It can be demonstrated that the detection results of expanded dataset generated by DCGAN in ResNeXt50 with FPN Faster R-CNN network is better compared to the original dataset and the Resnet50 with FPN Faster R-CNN network.