Research And Implementation Of Online Detection Algorithm For Fabric Defects

The development of textile industry greatly accelerates the economic development of our country. With the promotion of economic internationalization and market economy, the quality of textiles has become the key factor of affecting the competitiveness of enterprises. In the past 20 years, defect detection is still mainly relying on the manual detection, but this method can not guarantee the textile quality due to eye problems and environmental risks. With the coming of the information age, the computer is stepping into almost all areas; industrial automation is more and more becoming a trend which cannot be halted. In addition, the research of fabric defect detection system based on machine vision can effectively achieve the quality control of textile, and control labor expenses. Based on the existing research results of domestic image processing technology, there is no method to launch a real-time, online fabric detection system. It is necessary to further study the online detection algorithm.In order to achieve on-line detection, the algorithms of selecting the most suitable system are needed, which are the most efficient and the most suitable algorithms. Detection algorithm is the key to distinguish normal and abnormal texture, and then use a number of criteria to extract defects in the region. Besides above algorithms, the detection algorithm also contains features extraction, defect classification and other steps to get information which can be used to characterize the texture, and then specify the classification learning algorithm to distinguish information.This paper mainly studies the following contents:1. Pre processing of the collected images. Research and analysis of all kinds of noise processing algorithms, reduce or even eliminate the noise caused by the system or image acquisition, transmission process. The research of threshold segmentation and image enhancement algorithms are used to make the image most close to the target requirements.2. An improved algorithm based on LSD is proposed for the study and realization of towel count. Edge is a symbol of the segmentation of the towel, the performance of a straight block. After determining the center point of the mark, the number of the towel is determined, and the defect location is determined. First, the algorithm uses a simple edge detection method and then straight line detect, to replace size reducing method to achieve the elimination of aliasing. It also requires gradient calculation and searching for the most long connected vector, and then post process latitude’s and longitude’s gray values, so that to get the center of the mark. Experiments show that the edge detection rate of this algorithm is higher than other algorithms.3. The detection algorithm based on template matching is proposed. According to the gray level change cycle in the standard image, the template matching method based on gray level is used to locate the basic element, and then the variable template is constructed by reducing the matching information and searching range. And compared with other algorithms, it is proved that there is a higher detection rate.4. A region growing’s post processing algorithm based on region segmentation is proposed, which is used to study the defect segmentation and feature extraction. After the seed selection, the whole area is covered by the growth, and then the region is merged, which is intended to merge spot, bulk defection, and can improve the accuracy of the segmentation and can extract feature easier. A large number of feature values need to distinguish, select the most appropriate to reduce the amount of computation. Compared with other post processing algorithm, this algorithm for segmentation has better effect.5. Focus on the mechanism of BP neural network and design it according to the characteristics of this thesis. A comparative study of the defect classification algorithm, especially the BP neural network structure, the model construction (feed forward algorithm, the back propagation algorithm), and then design and learn it.6. Development of software platform based on OpenCV. Software can connect the camera, and can handle video, to detect the number of towels, defect number, the number of towel where the defect places and the type of defects,which get a good effect in the online detection, and then through the optimization of the algorithm to achieve the best results.