Video image quality anomaly classification method based on GoogleNet

Abstract

The invention discloses a video image quality anomaly classification method based on GoogleNet, and relates to the technical field of video image quality processing, and the method comprises the following steps: obtaining original data information in advance, and carrying out the data preprocessing which comprises the steps: marking the data information, carrying out the data augmentation, splitting a data set, and generating a tfrecore file; establishing a neural network model and performing training, wherein the neural network model comprises a pre-training weight of a calibration Image Netimage data set, a convolution base of a calibration model and a dense connection classifier; and taking the trained neural network model as a video image quality anomaly classification model and outputting a result. The method is simple to implement, high in classification accuracy, high in speed, capable of detecting image quality anomalies in real time and classifying the anomalies, convenient to operate and convenient to expand.

Description

technical field [0001] The invention relates to the technical field of video image quality processing, in particular to a method for classifying video image quality abnormalities based on GoogleNet. Background technique [0002] In the prior art, there are a large number of moving objects in the video images acquired by the video surveillance system, and among all the moving objects, two types of objects, people and vehicles, are generally the main attention objects. There are obvious differences in the management requirements of these two types of objects, so there is a classification requirement for these two types of objects in the video surveillance system. [0003] At present, the main methods in research and development are based on statistical training methods to classify objects. However, the use of such methods requires the collection of a large number of image samples of vehicles and people, and the recognition speed is slow, and the demand for computing equipment...

AI Technical Summary

Problems solved by technology

By using the contour features of the target to classify the target, the accuracy of the classification is improved; the size of the moving target is normalized by the scaling factor, which overcomes the defect of inaccurate width and height ratio features caused by the existing normalization method. And calculate the color histogram through the joint probability distribution, which reduces the amount of data in the color histogram, but it still has the problems of low accuracy and low efficiency

Images