A target positioning and extracting method, device and computer readable storage medium

Abstract

The application provides a target positioning and extraction method and device and a computer readable storage medium, the method comprising: extracting features of a multi-feature SAR image by using a multi-layer backbone feature neural network, and then determining at least one of a first loss function value, a second loss function value and a third loss function value by at least one of a positioning head neural network, a semantic segmentation neural network and a semantic completion neural network; adjusting at least one of the multi-layer backbone feature neural network, the positioning head neural network and the semantic segmentation neural network by back propagating at least one of the three loss function values; and positioning and extracting a target region in the SAR image until the entire deep neural network converges. By extracting oil spill saliency difference features and constructing a multi-feature SAR image, the application can simultaneously realize positioning and semantic segmentation of an oil spill region, and reduce the influence of oil spill similars and speckle noise on the detection result.

Images