Agricultural land semantic segmentation method of remote sensing image

Abstract

The invention discloses an agricultural land semantic segmentation method for a remote sensing image. The method comprises the following steps: selecting an anchor sample in a positive sample pair with a closer distance in a feature space as a reference anchor sample; calculating the similarity between all the negative samples and the reference anchor sample, and judging the negative sample with the highest similarity with the reference anchor sample as a possible false negative sample; taking the part of true negative samples in the possible false negative samples as model-dependent true difficult negative samples in remote sensing image self-supervised contrast learning; designing false negative sample confidence calibration loss; training a neural network model by using the labeled sample, performing back propagation after the prediction of the neural network model and the real label of the data are lost, and updating the parameters of the neural network model; and inputting the remote sensing image into the trained neural network model, and outputting a semantic segmentation result. According to the method, false negative sample self-judgment is added, and an unsupervised or negative sample construction method is abandoned, so that approximate judgment and calibration of the false negative sample are realized.

Description

technical field [0001] The invention belongs to the technical field of remote sensing, and in particular relates to a method for semantic segmentation of agricultural land of remote sensing images. Background technique [0002] Agricultural land is the land directly or indirectly used for agricultural production, also known as agricultural land, including cultivated land, garden land, forest land, pasture land, aquaculture water surface, water surface of ponds, farmland water conservancy facilities, as well as field roads and all other agricultural productive properties. Land occupied by buildings, etc. In order to protect agricultural land, images collected by remote sensing satellites are needed to identify agricultural land and prevent the use of agricultural land from changing. As a potential learning paradigm, self-supervised learning can realize autonomous learning of invariant features of remote sensing images in a label-free manner. The basic principle of the exist...

AI Technical Summary

Problems solved by technology

Considering that the false negative samples are essentially the existing positive samples in the data set, although this type of method completely avoids the generation of false negative samples, it also means that the model will not be able to use the existing positive samples in the data set, which reduces the The ability of the model to extract invariant features

Images