The Research On The Seeds With Broken Glumes Of Hybrid Rice Based On The Computer Vision Technology

It is a normal phenomenon that the glumes of hybrid rice seed can not bewell-closed, the seeds with broken glumes are more difficult to be stored than general ricespecies. In the process of storage, the quality of seed with broken glumes is easy to go bad,which can also affect the quality of other normal seeds; therefore, we must find a shortcutto preserve them separately. It is difficult to realize the separation of large number of riceseeds according to artificial separation, so we must find a method which is moreconvenient and a separation instrument which is easy to be used. This research utilizes thecomputer vision technology separates and recognizes the hybrid rice seeds with brokenglumes, and solves the problem which was difficult to preserve those seeds with brokenglumes separately. The main content of this research include 4 processes: imageprocurement, the low layer of image processing, feature extraction and pattern recognition.(1) Designing computer vision technology hardware equipment. discovers the best gainpicture method through the massive experiments, many grain of ways laying asidepicture is easiest to divide the recognition under the white background the picture.(2) Carrying on underlying bed processing to the gain picture, found easily in theunderlying bed processing process to divide the picture the method is the split-blipthreshold value division law, after the division picture can achieve easiest to understandand the analysis effect.(3) Analyzing the morphological structure of normal rice seeds, the rice seeds with littlebroken glumes and broken glumes, picking out the shape parameter. In addition,picking out 6 Characteristic values as the Characteristic values of this experiment;these six Characteristic values include acreage, perimeter, macroaxis, brachyaxis,axial ratio and circularity.(4) Designing a three-layer BP neural net structure, input layer has seven neurons, outputlayer has three neurons, and hidden layer has two layers; and in the hidden layer, therehas seven hidden node points. The rate of discrimination to three kinds of hybrid riceseeds are 96%, 82% and 95% respectively by using BP NN.