Research On Grain Water Content Detection And Impurities Recognition Method Based On Machine Vision And Hyperspectral Imaging Technique

It's significant to improve the technical content of agricultural products and value-added products, increased commercialization of agricultural products in China. Machine vision and hyperspectral and other automated testing technology into the food quality testing, so that food quality has been accurate and fast detection, is an effective way to achieve one of the target. In this paper, where food particles and impurities as the research object, the machine vision and spectral identification technology for food grains and other impurities in the basic theory and application of methods, main contents and results are as follows:(1) Research on the hyperspectral feature of the grain. Collected still contained 956 bands hyperspectral image of four rice varieties divided into 9 categories in city JiaXing and ChangXing. By processing of hyperspectral data, PCA and ANOVA of 4-facts-5-levels orthogonal experimental analysis, optimization of halogen lamps which has the band length between 620-680nm, and the intensity between 6000-12000 lux as a light source, black epoxy boards as a background in the design of machine vision systems for impurities food grains.(2) Building grain water content prediction model based on their hyperspectral image analysis.Rehydration experiment is designed for hyperspectral image acquisition of maize 02102 and wheat NS21 under different water content conditions between 10% to 20%. MSC spectral data pretreatment is used firstly. Through Linear Regression get 4 and 5 Band values for corn and maize respectively which have the biggest R2 with water content. Then establish a 4-5-1 and 5-5-1 three layers BP ANN for corn and maize respectively by empirical formula to predict the moisture content. The R2 of Regression Analysis in corn and maize are 98% and 96% respectively.It illustrates that hyperspectral imaging technology for corn and wheat moisture content non-destructive detecting is feasible. It establishes the basis of a new method of grain moisture detection and removal of water for high spectral detection of grains.(3) Machine vision hardware system and recognition model for the identification of grains and impurities is built.According to national standards GB1351-2008, samples are divided into six sub-categories. Choosing a white paper for imge acquisition background using the threshold segmentation method.On the basis of a large number of experiments (6400 images), using 29 characteristics of the sharp or color of the image for modeling the pattern recognition. First through a decision tree recognition method based single feature identified 68.6% of impurities. Then established a 28-9-2 three layer BP ANN recognition model, using L-M optimization as network training algorithm, logsig and purelin as hidden layer and output layer activation functions. The results show that the model's overall recognition rate of 90% or more. Finally it lists the main reasons for causing false positive model like incorrectly classified, low resolution compared to the object and covering the feature during image grab processing.