Research On Quill's Defects Detection Based On Computer Vision

China is the biggest badminton producer, accounting for more than 50% the world's production. Badminton production process has not been changed for 150 years since it was invented, this semi-manual processes are time consuming, labor-intensive and unreliable. With increment of labor cost, companies must improve technological innovation of manual intervention; therefore the research of badminton automated production process has great practical value. Some researches about the automation processes of badminton production have been done by some institutions, in fact, their results mainly focused on measuring size and detecting color of feathers, very few studies were done on defect detection. However, defect detection is a key point to badminton quality, especially rod defect of feathers, it relates to the resistance of playing badminton, defect feathers should be detected at the beginning of production.Based on machine vision inspection of quill defect detection, the process is to fix the feather at the station in specific direction, and capture feather images via cameras at different positions of station. After comprehensive analysis of the image processing, quill can be extracted, then defects can be detected in the hair, the final test results will be sent to the feathers delivery system, to complete the classification of feathers.Manufacturer need to address more than two million feathers every day, the speed of automation is the first consideration. To achieve this purpose, this paper has done some researches on feather rod defect detection with limited image pixels. Study found that defect detection have difficulties in the following:first, due to structure and size, some defect images are not clear under the normal image acquisition environment; Secondly, color difference is very tiny among rods, feathers and hairs, meanwhile width limit by rods is not significant, it is difficult to determine the threshold value; Thirdly, the uplift surface and the edge effect image of hair and rod, makes consistent shadow of gray in some areas. Currently the imbalance of gray shadow has no general method to compensate, at this point it will undoubtedly increase the difficulties of defect detection; Finally, the background of defects is quill, it must be identified before defect detection, the identification rate of a single feather is the product of two successful operations. To ensure final identification rate, every single step of the process should go with the best results. In this paper, starting from optical imaging analysis, studies have been done in pretreatment process, quills process, and defect identification system, the major research and innovation summarized as follows:1. Built up a full-automation feahter test platform which include feather thickness test, tortuosity and camber test, square and roundness test, colour test and defect test. Decressed the taches in test each characteristic separately. The platform included feather transmission system, functions of which contained feather selection, orientation, camera shooting and transmission; the image collection system, the function of which contained light source setting, type and model selection of camera, orientation and control; the image processing system, the function of which contained get the images from the image collection system, analyse, process and identfy the images and transmitted the final result to the transmission system to sort the feather.2. Combine with the physical structure of the quill, analysed the imaging process of the feather and the quill with PHONE theory. Posed a method using facelight and sidelight image select method to test the defect, which can extremely improve the imaging rate of defect. Focuse on the distribution of brightness nonuniform phenomenon built up a collection system illumination model. Deduced the brightness distribution function form the model. The brightness of image has been compensated to make it basicly balanced.3. Posed a dynamic active model base on centre line model to collect the quill. This algorithm base on centre line's symmety, use the centre line and the width to describe the outline of the quill. The 2D outline curve of model has been simplifyed to two independent 1D function. Using the continuity and derivability enegy to indicate the inner enegy of the model. The improved algorithm can avoid the strong outline disturb effectively, decrease the quantity of compute. Combine with the initial outline automatic setting algorithm, no intervention division of the quill can be realized. 4. A method ehich using the coupling impact filtering complex spread function to filtering the image has been posed in the paper. In this method the impact function can sharpen the border. The real part of complex spread can restrain the noise. The imaginary part can test the border. The tow part coupled each other, overlaid complex spread with appropriate impact function, based on the impact function which depend on the strength of the border to choose effect of strengthen or reduce. Bacauce the border strength of the defect is higher than that of noise, the defect will be strengthened and the noise will be reduced.5. Considered the Quantum error of image background will cover the characteristic of defect border, posed an improved ridge wave algorithm. This algorithm decreased the gray level difference between the quill and the background, weaker the effect of the quill border Quantum error, drew the linear characteristic of defect, reduced the range of defect direction, improved the test efficiency, can detected small defect of different angle, was propitious to raise the identify rate of the defect.