Object Detection Method of Optical Remote Sensing Image Based on Multi-scale Feature Learning

Abstract

The invention discloses an optical remote sensing image target detection method based on multi-scale feature learning, which mainly solves the problem of insufficient shallow feature extraction in the prior art, uneven distribution of target features of different scales, inability to train parameters as a whole, and the need for a large number of intermediate results. Dumping causes the problem that network training takes a long time. The specific steps of the present invention are as follows: (1) construct a multi-scale feature network; (2) construct a training sample set and a training class label set; (3) obtain the deep and shallow features of the multi-scale feature network; (5) Construct a test sample set; (6) Test the test sample set. The invention has the advantages of richer shallow detail feature information extraction of small targets in optical remote sensing images, high target detection accuracy and fast training network target detection speed.

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 multi-scale feature learning in the technical field of optical remote sensing image target detection. The invention can be applied to identify and detect targets in different areas of optical remote sensing images. Background technique [0002] With the continuous improvement of image spatial resolution, target detection technology based on high-resolution optical remote sensing images has been more and more popular and concerned. High-resolution optical remote sensing images have the characteristics of high spatial resolution, wide coverage, and convenient acquisition methods. The objects in the images retain rich color and texture features, so they can be widely used in many fields, such as military fields, road planning and Traffic supervision, smart city construction, environmental monitoring, ag...

AI Technical Summary

Problems solved by technology

This method can accurately and richly represent the features of the target, and can fully extract the target candidate frame and other advantages. However, the method still has the disadvantage that the information of the small target needs to go through six convolutional feature extraction layers to extract features. , shallow features are filtered out, and the distribution of target features at different scales is uneven, so the detection accuracy for small targets is low

This method reduces the calculation of features by preprocessing the image, and then uses the regional convolutional neural network to extract the deep features of the image, which can better represent objects, and classifies the features with the support vector machine classification algorithm, and obtains good results. However, the disadvantage of this method is that since the network training process is separated, extracting candidate frames, extracting image features, and using support vector machine classification for independent training, a large number of intermediate results need to be dumped, which cannot be trained as a whole parameters, so network training takes a long time

Images