Dense target feature learning-based target detection method of optical remote-sensing image

Abstract

The invention discloses a dense target feature learning-based target detection method of an optical remote-sensing image. The method mainly solves the problem that small-target information is filteredout due to deep convolution in the prior art. The method comprises the following specific steps: (1) establishing a dense target feature network with a total of 25 layers, and setting parameters of each layer; (2) constructing a training sample set and a training class label set; (3) acquiring the deep and shallow features of the dense target feature network; (4) fusing the deep and shallow features of the dense target feature network; (5) obtaining a target candidate frame feature set; (6) carrying out dense pooling; (7) constructing a test sample set; and (8) detecting the test sample set.The method has the advantages that extraction on the deep and shallow features of the optical remote-sensing image is good, and accuracy of target detection is high.

Description

technical field [0001] The invention belongs to the technical field of image processing, and further relates to an optical remote sensing image target detection method based on dense target feature learning in the technical field of optical remote sensing image target detection. The invention can be applied to identify and detect ground objects in different areas of optical remote sensing images. Background technique [0002] Object detection technology is one of the core issues in the field of computer vision. Remote sensing object detection uses images captured by remote sensing satellites as data sources, and uses image processing technology to locate and classify objects of interest in the images. Remote sensing target detection is an important part of remote sensing application technology. It can capture attack targets and provide accurate location information in high-tech military confrontation, which is of vital significance in the military field. [0003] With the f...

AI Technical Summary

Problems solved by technology

This method can accurately and richly represent the features of the target, and can well extract the target candidate frame, and reduce the extraction time of the candidate frame by sharing the convolution parameters. The information is filtered out after convolution layer feature extraction and ROI pooling in the region of interest. Only larger-scale targets can be detected, and the detection accuracy for dense small targets is low.

This method assists the detection of remote sensing image features by training a classifier to separate sea and land, which can predict the results of robust target detection and reduce the missed detection of ships caused by side-by-side placement. However, this method still has shortcomings. The disadvantage is that due to the need to use multi-layer convolution to obtain feature maps and predict targets with deep convolution, the multi-scale features of the target are not considered, resulting in insufficient utilization of target information, resulting in unsatisfactory detection of dense small targets

Images