Study On Appearance Quality Nondestructive Inspection Of Peanut Using Computer Vision

There are definite stipulations on the imperfect granularity, musty granularity, size, shape and other apparent quality of peanut in the export.But at present, there is only a kind of photoelectric device which can pick out the musty granularity based on the color in the market, and other apparent quality can only be identified manually, which can not satisfy the need of the market.Computer vision technology has the features of lossless and rapidness, and it can test several index of quality at a time, it is also convenient for designing and manufacturing production line which can classify agricultural products automatically. All of these show us a bright prospect of the technology on agricultural products auto identification. Using computer vision technology to research the apparent quality of peanut and realizing the apparent quality test of peanut automatically without losses make great sense on enhancing the competitiveness of our peanut. Against such a background, a method of apparent quality test of peanut is discussed in the paper, a designed system suitable for analyzing and testing the apparent quality of peanuts is also included. The main contents are as follows:1. Basing on the practical demand, several commonly used color models are analyzed, and the basic image processing algorithms of sound filtering, image enhancing, image dividing, and feature extracting are also analyzed and studied. Finally, a basic image processing algorithms suitable for the apparent quality is chosen.2. To realize the test of musty peanut, the change law of the parameter of feature of color and the parameter of feature of vein is studied in the process of the peanut becoming musty. The feature of color H,I,S and the feature of vein RW,GW,BW are extracted as the inputs of Neural Networks created by MATLAB, and the identification of normal peanut, stale peanut and musty peanut is realized, with the accuracy rate of 95%,90%,100% separately. To identify the musty extent of peanut, H and S Threshold is used to identify the area which is musty. By processing the area morphologically, calculating the area ratio of musty area by total pixels, the musty extent of peanut coat is identified, with the accuracy rate of 90%.3. To realize the test of imperfect peanut, the feature of color of the imperfect area is extracted, and the system of peanut imperfection test with the parameter of feature of R, G, and B color information is established based on the match of mode, and the identification of perfect and imperfect peanut is realized, with the accuracy rate of 88.56%.4. To realize the test of different shapes of peanut, Fourier Descriptor and inverse Fourier Descriptor is used to describe the shape of peanut, it is found that the feature of fist thirteen Harmonic Changes can represent the main shape of peanut in the Fourier Descriptor. The Fourier Descriptor and the Artificial Neural Network realize the test of the shape of peanut of oblong, simple, Trilateral, Elliptical and circular, with the accuracy rate of 90%, 93.3%, 96.7%, 100%, 93.3% separately.5. To realize the test of different sizes of peanut, the method of extracting the geometrical parameter of area, circumference, the length of major axis, the length of minor axis, roundness, eccentricity, equivalent diameter and the compactness of peanut is studied, and the correlation model of projected area and the weight of peanut are established. The result shows that there is a significant relativity between projected area and the weight of peanut. Based on the network system with the support of vector machine and geometrical feature parameter, the accuracy rate of the identification of the size of the peanut with five classes is 90%,95%,95%, 90%, 100% separately.6. To test the shelf life of peanut by the vision technology, the change law of the color of the coat, the vein and the shine of peanut is studied, and the Bays classifier is used to establish the model which describes the relation between the three parameter of feather of color of H, I, S as well as the vein of RW, GW, BW and the time of storage, and the accuracy rate is 88%.7. The design of the hardware and software system including machine, vision test, control and program provide the foundation of the rapid test of peanut without losses.