A Computer Vision System For Real-time Hair Impurity Detection In Mushroom Semi-product

Currently, detecting non-metallic impurities such as hairs in mushroom semi-product was still mainly depending on artificial visual inspection. Therefore, the inspector's subjective judgments, experience, testing technology and other factors will affect the objectivity and accuracy of the test results. Moreover, manual inspection was time-consuming, laborious and inefficient. Therefore, we needed high-speed, objective and accurate inspection methods to find hair in juicy mushroom automatically. This was exactly the main purpose of this research project:to achieve an on-line hair impurity test system for mushroom production.The real-time inspection system detected the hair in mushroom using digital image processing and feature extraction techniques. This system helped to reduce labor costs and improved the detection efficiency greatly. This study focused on image feature extraction and target recognition based on computer vision technique. After a survey on the related research work, we explored the feasibility of hair impurity recognition in mushroom juice. The main contents of this article were as follows:(1) Set-up of image acquisition system to obtain high-quality mushroom semi-product image. According to the requirements of subsequent image processing algorithms, a machine vision system was designed and an experimental platform was developed for taking high-quality images with an appropriate camera height and light source. After several experiments, the DH-SV2000FC/M digital camera produced by Daheng Limited cooperation was selected as image acquisition devices. We used dome lighting as light source in our experiments.(2) A series of foundational image processing algorithms, such as image graying, image denoising, image segmentation, edge detection operators and morphology calculus were analyzed in this study. The hair impurity identification system algorithm was developed by comparing the performance of different image processing algorithms in each stage.(3) The core algorithm of hair impurity detection was linear structure detection algorithm. Firstly, we studied the three most popular linear structure detection methods:Morphological Method, Double Edge Method and filter Function Kernel Method. Then a linear structure detection algorithm combining local gray value and continuity of the linear structure was presented in this work. Experiment results proved that this linear structure detection algorithm have excellent performance.(4) The parameter selection problem was studied for the proposed linear structure detection algorithm. Firstly, we detected the direction of linear structure on eight different directions. Secondly, we judged the direction continuity between the current pixel and the adjacent pixel. Finally, the ultimate hair impurity detection results were determined.(5) The hair impurity detection algorithm together with a user interface were implemented using Visual C++, OpenCV and Daheng software development kit. Various experiments were carried out to evaluate the system performance.The experiment results showed that the real-time hair impurity detection system proposed by this paper was practical and having a broad market prospect.