Weak supervision remote sensing target detection method based on hybrid hole convolution

Abstract

The invention provides a weak supervision remote sensing target detection method based on hybrid hole convolution. According to the method, various customized designs such as hybrid hole convolution,attention mechanism and multi-layer pooling are adopted, multi-scale feature extraction and fusion are enhanced, and the robustness of objects of different sizes is improved. Besides, asynchronous iteration alternate training between a strong supervision detector and a weak supervision detector is utilized, training and detection can be carried out only through an image-level real label, and the purpose of cooperatively improving the detection performance is achieved.

Description

technical field [0001] The invention relates to the field of pattern recognition, in particular to a weakly supervised remote sensing object detection method based on mixed hole convolution. Background technique [0002] With the development and combination of aviation technology and computer vision technology, high-altitude high-resolution optical remote sensing images are becoming easier to obtain and are used in various fields. As a basic feature extraction problem in remote sensing image analysis, the research on this field has a long history in academia. Specifically, the targets of remote sensing image target detection include the location of ground objects and the classification of object categories. In recent years, research results in the field of remote sensing image target detection have progressed rapidly, and many algorithms can simultaneously achieve high-precision ground object positioning and recognition. Among them, most of them decompose the image feature...

AI Technical Summary

Problems solved by technology

However, this type of method generally predicts multi-scale feature maps separately, and the network will become very complicated and not conducive to training

[0005] In addition, another important problem in the field of remote sensing image target detection is the lack of labeled data sets. The advancement of remote sensing technology has brought a large amount of high-resolution data, and these image data also contain a large number of target objects to be detected. It takes a lot of manpower and material resources to mark the detection frames of objects in the image one by one

Images