Application of improved MobileNet-SSD on underwater sea cucumber detection robot

We present an underwater sea cucumber detection method based on improved MobileNet-SSD (MD-SSD), which is used to improve the serious loss of accuracy of MobileNet-SSD in sea cucumber detection. By combining the residual structure with the dilated convolution to improve and optimize the basic network MobileNet. Firstly, the depth separable convolution in the network is replaced by the two residual units designed in this paper, and then the context information is performed by using the dilated convolution in the convolution layer of the first two residual units. Tested on the data-enhanced sea cucumber dataset, the accuracy and recall rate of the MD-SSD model reached 93.55% and 92.68%, which is 2.76% and 6.29% higher than that of MobileNet-SSD. The detection speed is 43.65 frame/s higher than that of SSD, detection accuracy is higher than SSD 0.46%. In addition, the lightweight network MD-SSD is applied to the remotely operated vehicle platform, which can accurately and rapidly detect sea cucumbers under the CPU environment in the self-built datasets and simulated underwater environment.