Semi-supervised Remote Sensing Image Object Detection and Segmentation Method Based on Class Activation Map

Abstract

The invention requests protection of a semi-supervised remote sensing image object detection and segmentation method based on a class activation map. First, use the given remote sensing image annotation data to generate a classification annotation data set, train a global average pooling GAP (Global Average Pooling) classification convolutional neural network model, and use the principle of weight superposition of feature maps to construct a classifier that can generate classes The convolutional neural network model of the activation map CAM (Class Activation Mapping); then, after data enhancement, the class activation map and the real label are used as the training targets to conduct semi-supervised training on the target detection and segmentation models; then, using the real label The test set is used to verify the target detection and segmentation model, and a model with high detection and segmentation accuracy is obtained; finally, when only a small amount of labeled data is used for training, the method has a good remote sensing image target detection and segmentation effect.

Description

technical field [0001] The invention belongs to the technical field of image target detection and segmentation, in particular to a semi-supervised remote sensing image target detection and segmentation method based on a class activation map. Background technique [0002] Convolutional neural networks (CNNs), as a model architecture for deep learning, have become the most effective method in the field of image processing and computer vision. The two characteristics of weight sharing and local receptive field reduce the number of weights, which reduces the computational complexity of the model; the translation invariance of image features also makes it have good feature extraction ability and high stability. [0003] At present, a large number of studies have used convolutional neural networks to explore target detection and image segmentation methods. Since R.Girshick et al. proposed the R-CNN deep learning model based on candidate regions in 2014, a number of classic target ...

AI Technical Summary

Problems solved by technology

[0005] For target detection and segmentation of remote sensing images, due to the large amount of data, the labor cost of data labeling has also increased significantly. Existing datasets usually only have relatively easy labeling of target frames, while pixel-level segmentation labeling is rare. less

Images