Salient target detection method based on multi-scale expansion convolutional neural network

The invention discloses a saliency target detection method based on a multi-scale expansion convolutional neural network. The method comprises the following steps: extracting multi-scale features of an input image; inputting the multi-scale features into an expansion residual error convolution module to obtain fusion features including context information of the multi-scale features; respectively inputting the fusion features into a plurality of channel attention modules to obtain a plurality of saliency features; and carrying out dimensionality reduction activation on each display indigenous feature to generate a saliency map, and carrying out deep supervision training by adopting a mixed loss function fusing cross entropy and intersection-union ratio loss. According to the method, rich global semantic information and local semantic information in the image are fully captured by using the expansion residual convolution module based on the multi-scale expansion convolutional neural network, the problems that