A Novel Fish Counting Method with Adaptive Weighted Multi-Dilated Convolutional Neural Network

With the rapid development of computer technology, combining computer vision and artificial intelligence technology with traditional aquaculture industry has become a new development field. Fish density estimation can provide strong support in many fields such as feeding density control, fishery resource survey, and bait feeding. The traditional target recognition based counting methods have problems such as large error and low efficiency, which cannot be applied to high density fish counting. In order to solve the problem of traditional counting method, this paper presents a novel framework (WMD-CNN) which can estimate the density of fish accurately. The proposed framework consists of three parts: the simplified VGG module, the multi-dilated convolution module and the squeeze-excitation module(SE). In order to verify the effectiveness of the proposed algorithm, this paper has done a lot of simulation experiments. Experimental results show that the proposed method has good robustness and stability.