Sliding window scanning method based on margin cutting method

Abstract

The invention provides a sliding window scanning method based on a margin cutting method, and belongs to the technical field of graphic processing. According to the technical scheme, the method comprises the following steps that step 1, inputting an original large image; step 2, cutting allowance of the large image; step 3, performing conventional detection on subgraphs; and step 4, performing unified annotation and log reduction; step 5, completing the whole scanning process. The method has the advantages that the large image is divided into the sub-images of the same size to ensure that thesubimages are not damaged, and meanwhile a small target is not distorted. The small target is amplified in a phase-varying manner in the subgraph. The detection precision of the detection difficulty target is greatly reduced. An intermediate result of the subgraph detection is stored. Unified marking according to logs in the statute stage is performed, so that although cutting detection is performed, no extra internal storage is consumed, and the subgraphs are detected according to the sliding sequence of the time window, so that the graph taking mode can ensure that the target detection is not heavy and is not leaked.

Description

technical field [0001] The invention relates to the technical field of graphics processing, in particular to a sliding window scanning method based on a margin cutting method. Background technique [0002] The traditional target detection method directly takes the original picture as input, and the deep learning model extracts the features of the picture through the convolution algorithm, classifies and labels the picture, and takes the marked picture as the output. The output picture not only has target category information, but also includes the specific coordinates of the target category in the picture. [0003] The detection model generally only processes images of fixed size (resolution). Although the original input image size is different, the detection model is generally processed by resize to shrink / reduce it to a fixed size. If the original image If the size is large and contains small targets, the pixels of the small targets will be lost too much after resizing an...

AI Technical Summary

Problems solved by technology

[0005] The present invention aims at the above existing problems, in order to overcome the defect that the small target is distorted after being resized and the small target is easily cut due to arbitrary segmentation, the present invention proposes a sliding window scanning method based on the margin cutting method

Images