Target detection and recognition method and device for high-resolution image

Abstract

The invention relates to the technical field of image recognition, in particular to a target detection and recognition method and device for a high-resolution image, and the method comprises the steps: obtaining a high-resolution image set, carrying out the data enhancement, and obtaining an enhanced image set; segmenting each original image in the enhanced image set to obtain corresponding sub-images and position information thereof; performing coding and fusion processing on the sub-images and the position information thereof to obtain corresponding data tensors; based on a Faster R-CNN model, performing layer-by-layer feature representation learning on the data tensor, fusing low-layer information, middle-layer information and high-layer information of the Faster R-CNN model by adopting an attention mechanism, determining feature representation corresponding to a sub-image, further determining a candidate target position, and performing regression and classification so as to determine a final target position and a category to which the final target position belongs; and determining the final target position and the category of the original image according to the final target position and the category of the final target position. Through the scheme, the final model performance is improved.

Description

technical field [0001] The present invention relates to the technical field of image recognition, in particular to a method and device for detecting and recognizing targets in large-resolution images. Background technique [0002] With the rapid development of information technology, the convenience, efficiency, safety, and reliability brought by information processing have made industrial informatization a development trend in all walks of life. As the most ubiquitous medium in daily life, images play a key role in the process of information transmission. Therefore, how to efficiently and reliably use image information as one of the important research contents in the direction of computer vision has attracted many researchers. [0003] In the early days, due to the complex semantic information of images, traditional machine learning algorithms could not fully understand image information, so the research done was relatively simple. In recent years, the emergence of deep l...

AI Technical Summary

Problems solved by technology

Taking the Landsat image as an example, when the image is used to observe terrain, vegetation or water conservancy, it needs to detect its target, and if the target is small, because the image resolution is too large, the ordinary image target detection method will not be able to It works. First, the input of ordinary methods is generally located at 10 2 *10 2 - 10 3 *10 3 magnitude, and the size of a large-resolution image is often much larger than this size. If the size of the original data is simply scaled, a large amount of information will be lost, especially when the detection target is small, and the target may even be lost; Second, large-resolution images are larger in size and rich in information, making the ratio of the target area to the background area smaller; third, due to its special application background, the amount of such images is small and cannot be obtained A large amount of experimental data is not conducive to the training of the model

Due to these factors, the original method cannot obtain good accuracy on the problem of large-resolution image target detection, so it cannot meet the normal performance requirements

Images