Water surface garbage identification method based on deep learning

• Main Authors: YUAN BAOHUA, YANG QINGYUAN, XU SHOUKUN, GONG YAN, LI NING, ZHANG TINGBO, YUAN HAOMIN
• Format: Patent
• Language: Chinese; English
• Published: 13.01.2023
• Subjects: CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; PHYSICS

The invention relates to the technical field of image processing, in particular to a water surface garbage recognition method based on deep learning, which comprises the following steps: cutting a water surface garbage recognition data set picture; adding the feature graph and the input graph through residual connection to obtain a suppression weight; multiplying each channel of the input graph by the subtracted suppression weight element by element to obtain a weighted feature graph; the deep layer uses two depth separable convolutions to form a DSC module, and the shallow layer uses a Res module; obtaining a one-dimensional feature vector by combining depth separable convolution of a channel attention mechanism, then generating a channel weight by using one-dimensional convolution, and finally multiplying the channel weight by an original feature map; an ECA module is added behind the Res module and the DSC module of the encoder; the decoder uses a depth separable convolution module. According to the method