The Research Of Specific Object Recognition Based On Machine Vision

Badminton is labor-intensive product. It is some ten procedures from the feather picked out to badminton. Each procedure need equipment and human labors, so it cost plenty of space and labors to produce badminton, and the detection and the classification of feather are the procedures in which most people work. In modern industrial production and research, the aim to reduce the cost or avoid casualties, machine vision instead of the human eye detects or recognizes feathers in more and more places. However, the craft highly dependent on labor to produce badminton is not greatly improved for several decades. Therefore, the thesis designed a system based on machine vision to classify feathers.The thesis mainly researched the feather images from four views which are pre-processing, texture feature extraction, color judgment, pattern recognition. The innovation and achievements are as follows:1. Each feather has the feature which is different tiny texture and arch degree, which will influence texture feature extraction and classification. The thesis firstly use median filtering which radius is three to reduce the influence of tiny texture, and use homomorphic filtering weaken uneven illumination which is caused by feather uneven.2. According to the experimental data which got attained when Gray Level Co-occurrence Matrix select different step length and angle, it is analyzed that seven parameters such as contrast are as feather characteristics. Then an original method had been put forward to find the best step length and angle. That is to change one of the two variables, and to get the result of the variance and the difference of each parameter, then according to the result to select the best step length and angle. Those are 20 and 45°respectively in the thesis. In the pattern recognition experiments, the seven parameter values are as the input values of neural network. BP neural network is well to distinguish good feather from the other class.3. Firstly, It is used CIE LUV color space to get Euclidean distance between the feather to be measured and standard feather. The result showed that this color space is not suitable to classify the feather. Secondly, a new method based on HSI color space had been put forward. This method reduce the component I to color, which is that H, S and I get the weight 16,8 and 1 respectively. The experiment showed that HSI color space can be well to classify the feather color.4. Weak bug bite feather is the feather was bitten slightly. With the feather bitten serious different, it can not be judged by threshold segmentation, and can be mixed with noise. In this thesis, Canny algorithm was used to detect the defect of edge, a algorithm was added to take precautions against the event, using the Character of connected component extract the defect edge information in the end.The system was built on MATLAB platform. It can detect feather defects and classify color, which not only reduces the need in labor, but also improves the efficiency in production, and it has certain practical value.