Violation detection model generation method, computer equipment and readable storage medium

Abstract

According to a first aspect embodiment of the invention, the invention provides a violation detection model generation method, which comprises the steps of receiving a sample picture; extracting samples in the sample pictures based on pixel segmentation, and forming a training set by a plurality of samples; and training the to-be-trained model by using the training set to generate a violation detection model. According to the violation detection model generation method provided by the invention, after the sample pictures are received, the samples in the sample pictures are extracted based on the pixel segmentation technology, so that the samples are extracted according to the contours of the samples, and a training set is formed by a plurality of samples; the to-be-trained model is trainedthrough the training set, the violation detection model is generated, for articles needing to be transported or stored in a covering mode such as coal piles and sand piles, recognition is conducted according to the contours, background information can be prevented from being extracted, overlapping extraction of multiple samples is avoided, the risk of false detection or missing detection is reduced or even avoided, and the accuracy of the detection result is improved.

Description

technical field [0001] The present invention relates to the technical field of environmental detection, in particular to a method for generating a violation detection model, a computer device and a computer-readable storage medium. Background technique [0002] With the success of deep learning applications in the field of image and video object detection, more and more new technologies have begun to further improve the accuracy and speed of object detection. At present, target detection can be roughly divided into two categories: One-stage (single-stage) and Two-stage (two-stage). The former is represented by different versions of YOLO and SSD. Their feature is to extract candidate frames during the detection process. The same network is used to extract the features of the target, or the two are combined into the same process, so it is called One-stage; the latter, the Two-stage method, combines the process of extracting candidate boxes with the extraction The process of t...

AI Technical Summary

Problems solved by technology

[0003] However, no matter what kind of target detection method is currently used, there is a problem, that is, simply use a rectangular frame to specify the position of the target

In fact, when this detection method encounters more complex problems, such as irregular edges of targets, overlapping between targets, and similar situations where different categories are close to each other, the detection accuracy is relatively low.

The first case is easy to understand, because the rectangular frame often includes background information that does not actually belong to the target. The second and third cases are calculated when judging whether to merge between different rectangular frames and rectangular frames. When IoU (Intersection over Union is the evaluation function), the rectangular boxes of different targets are often merged together. These three situations will eventually lead to false detection or missed detection of the target.

Images