Binocular Vision On-line Detection System Study For Conveyor Belt Longitudinal Tear Based On Infrared And Visible Light

According to existing conveyor belt longitudinal tear on-line detection technique, the on-line detection method, signals acquisition of image feature,real-time processing of image, registration algorithm and early-warning strategy of the binocular vision detection system is deeply studied for conveyor belt longitudinal tear based on infrared and visible light in this paper in order to realize the conveyor belt longitudinal tear detection system with characteristics of automation, rapidity, robustness, precision, reliability and suited engineering application. Then the feasible detection system is developed for field application of conveyor belt on-line detection. The main study of this thesis is organized as follows:First, concentrating on binocular vision on-line detection theory based on infrared and visible light, we implement the key technical issues and relevant science issue of feature extraction and recognition, early-warning strategy, etc.on conveyor belt longitudinal tear. This paper present a binocular vision on-linedetection method which combined the detection of visible light with adaptive line source compensation and infrared imaging recognition based on feature used in conveyor belt longitudinal tear.Second, the detection system of the mine conveyor belt longitudinal tear based on infrared and visible light CCD is designed. We design each subsystem by adopting modular design, and realized on-line real-time acquisition of the conveyor belt longitudinal tear fault feature image by the binocular vision sensor based on infrared and visible light. It avoids the delay problem of signal centralized processing by designing the corresponding processor for each unit according to requirement. Moreover, it is convenient for installation and adjustment, maintenance and extension.Third, making a deep research on the method of corner points, edge and linear feature extraction in conveyor belt longitudinal tear image. Further, we present the method of synchronous acquisition by triggering each other of infrared and visible light detection and similarity with the weight sum, which can realize the characteristic integration for global variable and local component.Additionally, the algorithm of real-time extraction, on-line processing and recognition for the local feature is developed. Moreover, the mathematical model of fault identification is developed and analyzed by its theory, and the experimental results of numerical simulation and feature image processing are showed in this paper.Forth, the multi-resolution image registration method for corner points, edgeand linear feature extraction for conveyor belt longitudinal tear based on infrared and visible light is proposed to solve the accuracy and reliability of breakdown judgment for conveyor belt longitudinal tear and the visual of fault early warning.Fifth, we present the deep learning algorithm of sparse coding neural network which is unsupervised by investigating. Then the real-time early-warning method of the database for classification and recognition of fault feature image is established. The robustness and practicability in this article is tested by the industrial test. The system achieve the requirement for field test of on-line detection warning of the conveyor belt longitudinal tear. Therefore, the advancement and practicable industrial application value of this article is verified through the experiment.