RGB-D collaborative saliency target detection method based on prompt learning

Abstract

The invention discloses an RGB-D collaborative saliency target detection method based on prompt learning, and relates to the technical field of deep learning. Comprising a training step of obtaining a training set; constructing a collaborative significance target detection network; performing data enhancement on each original natural scene image and the corresponding depth image in the training set, and inputting the original natural scene image after data enhancement and the corresponding depth image as original input images into a convolutional neural network for training to obtain a collaborative saliency target prediction image; calculating a loss function value; a convolutional neural network classification training model is obtained; and testing steps are carried out. According to the method, features are divided into RGB features and depth features, feature fusion is carried out on different sizes, image group common feature information is extracted from the deep features, shallow detail information is fused, common saliency targets in a group of images can be accurately described, and the collaborative saliency target detection accuracy is effectively improved.

Images