DEEP CONVOLUTIONAL SPARSE CODING NETWORK FOR SALIENT OBJECT DETECTION IN VHR REMOTE SENSING IMAGES

In order to reduce computational redundancy and increase the speed of image analysis, Saliency Object Detection (SOD) is one of the outstanding methods for Very High Resolution (VHR) remote sensing image analysis. However, Remote sensing images (RSIs) have the characteristics of diverse spatial resolutions and cluttered backgrounds, leading to the direct use of SOD methods for natural scenes generally not achieving satisfactory results. In this paper, combining the advantages of Convolutional Sparse Coding (CSC) and deep neural networks, a deep CSC network model is proposed for SOD of RSIs. First, a CSC Block (SCSB) is constructed by combining the CNN component and the Soft Shrinkage Threshold (SST) function to fully extract the effective information of the image. Then, build a multi-level coding network by stacking multiple CSCBs to enhance the perception of multi-scale and detailed information of salient targets. Finally, multi-level features are integrated in a simple way, and the entire network performs supervised learning in an end-to-end manner. The experimental results on the RSIs data set show that the proposed network model is superior to other methods in both quantitative and qualitative performance comparison.