Research On Key Techniques Of Flaw Inspection For Badminton Feather

China is not only a world badminton power, but also the largest badminton producing country in the world. As the most important part of badminton, the quality of badminton feather directly affects the final product quality.Due to the damage of feather caused in its natural growth and post-processing, there may be various flaws in processed feathers. The flaws seriously affect the quality of badminton feather. The flaw inspection of badminton feather is one of the most difficult processes of badminton making, and its completion is mainly manual work at present. The manual inspection is not only highly labor-intensive and inefficient, but also affected by individual elements like worker’s ability and experience, which, therefore, tends to cause inspection error or inspection missing.In order to improve production efficiency and guarantee product quality, it is necessary to study the automatic inspection technology for badminton feather based on machine vision. Research on the subject helps to cut production costs, reduce the employment needs, and actively promotes badminton-manufacturing transition from labor-intensive industry to technology-intensive industry. Therefore, the study of this subject has important theoretical significance and practical value.At present, the technology of machine vision has been widely applied in some industries. In contrast, research on the technology of machine vision inspection for badminton feather started relatively late. The application of machine vision technology in the badminton manufacturing industry is still in the initial stage. Now, papers retrieved on badminton feather inspection based on machine vision are very few. Some of those focused on crease detection of feather quill and size measurement of badminton feather. Moreover, far fewer of those focused on flaw inspection.This paper focuses its study on badminton feathers, in which the research includes the hardware equipment and software system of the badminton feather flaw inspection system (BFFIS), illumination correction, image smoothing and denoising, flaw segmentation, flaw feature extraction, flaw classification and identification, etc. All these are helpful to the improvement of the automation level of equipment in badminton manufacturing. The main research results and innovation points of this dissertation can be expressed as follows:1. Due to the need of performance analysis of feather inspection algorithms, image database of defective feathers and the sample database of typical flaws have been built, which provides unified experimental samples for different researchers, and solves the problem caused by the difference of experimental samples and its influence on the accurate comparision between relevant algorithm performances.2. Due to the uneven illumination of feather image caused by curved structure of feather surface, an improved illumination correction algorithm based on the Retinex theory is proposed, which sharpens the image contrast, improves the image quality, makes the image details clearer, and provides a basis for accurate detection and classification of feather flaws.3. The characteristics of feather flaw are not clear because of the influence of image noise and feather texture. According to characteristics of badminton feather image, this paper proposes an improved anisotropic mean shift algorithm for image smoothing and denoising. Compared with several kinds of typical denoising algorithms, the experimental results show that objective indicators and visual perception of this algorithm are better. While the algorithm reduces image noise effectively, it achieves a better effect on feather texture smoothing, and retains the defect information as much as possible, which provides pretreated images with clear flaw goals for the subsequent processing.4. There are various types of feather defects, and their morphological variations are extensive. Variational level set image segmentation model has the advantages of accurate segmentation, unnecessary artificial participation, and is suitable for multiple objects segmentation. However, it also has some disadvantages, such as high iteration and low efficiency. Combined the characteristics of pretreated image with the need of flaw segmentation, an improved C-V level set model is proposed. On the premise of meeting the demand of image accurate segmentation, the improved algorithm has reduced the iteration number of contour evolution, saved the running time, and advanced the efficiency of C-V level set algorithm.5. According to the characteristics of feather flaws, feature extraction methods are presented for color feature, shape feature, and texture feature of flaw area. Respective recognition methods are proposed for the four common types of badminton feather flaws. In this paper, different kinds of features are extracted respectively, and then mixed feature vectors are constructed according to the needs of flaws identification. This method can keep the feature vector at a lower dimension while identifying the flaws accurately.Combining the demand of feather flaw inspection with the research achievements of image processing field, this paper systematically puts forward the solutions to some key issues of flaw inspection, such as illumination correction, image smoothing and denoising, flaw segmentation, feature extraction. Simulation experiments have been conducted, and some good experimental results have been obtained. All of these provide a theoretical basis for the development of badminton feather flaw automatic inspection system based on machine vision. This dissertation has a certain practical value.