Research On The Appearance Quality Detection System Of Soybean Based On Image Processing

Soybean is one of the most important cash crops. Soyabean industry has permeates through every aspect of the society and influences the life of people. However, as an important part of soybean production and processing, the detecting techniques is traditional manual method.So it has seriously restricted the product quality and production capacity. This research is based on the exterior character and interior character contacts; we design a detecting systerm for inspecting soybean exterior quality based on image procession. It can be a new detecting technique with precise and high efficiency.Considering the differences between kinds of light and backgrounds. We have designed an black box as surroundings. CCD, computer and lux meter are used to build the hardware platform of the system, assembly language is used to compile the systems homologous software.The system capture images of nomal soybean, soybean frogeye, mildewed soybean,worm-eaten soybean, fractured soybean and mixed soybean. There are 16 pieces of soybeans every picture. To make the data precise, there are thousands of picture for the research.We take RGB color model and HIS color model to compare the outcome in the research. The research enhancing target-background contrast using grey linear transformation chose image thresholding based on amplitude segmentation. It carried on method of hole filling and overlapping segmentation to segment the binary images. The research featured parameter extraction of morphological, color and texture characteristics. They included circularity, compaction, eccentricity, bulkiness the shorter and longer axis in the ellipse and so on. Three-layer BP artificial neural network structure was designed using Matlab software. Compared with kinds of optimization functions, principal component analysis was used to raise training accuracy. Nodal points of hidden layer were selected to be 12 for one kind of defective soybeans detecting.The average recognition rate is 90.33%.An identification software system, which was complete and independent, was designed with VC++ and Matlab. It is easy for operation and humanized.At stated previously, the research provides a new kind essential theory foundation and technology support for soybean nondestructive measurement. The studied methods are signigicant and have great practical value for strengthning soybean competition in the international markets and improving the development of export trade.