Study On Application Of Machine Vision In Agricultural Product Detection

China is a large agricultural country, but China's per capita cultivated land is far below the world average, and has a low labor productivity in agricultural production and the follow-up processing. Quality detection of agricultural products is to ensure the quality of agricultural products and the key to the safety of agricultural products, and quality detection and control is often based on component analysis. Whether by chemical or instrumental analyzing, the samples pre-treatment and time-consuming of the experiment are not suitable for many occasions. Automatic detection of agricultural can greatly enhance the labor productivity. In detection of agricultural products surface detection is the most important, and machine vision technology for agricultural products detection has great advantages. The thesis for the practical needs of rice processing plants, based on the MATLAB software platform, build a computer vision-based recognition system of the rice quality; firstly, choices Image pre-processing algorithm; and then extracts the color features of rice image; lastly classifies rice by linear discrimination function and neural network based on the features. The main work is as follows:①Studies the feature extraction method of rice image. According to the features of rice image, the rice image recognition should be based on color features.②Analyses some pre-processing algorithms of image, such as image transform and image enhancement, then designs the approach of rice image pre-processing. Then rice images are transformed to HSI color space, and their H-component histogram are obtained; lastly extracts rice image features by principal component analysis and genetic algorithm, respectively. An improved criterion of within-class and between-class distance is used for determining the genetic algorithm fitness function.③Classifies rice images with linear discriminant function and BP neural network.④The experimental results are analyzed, and the thesis proposes a method of determining the final results by vote. experiment shows that the method can improve the classification accuracy rate。Finally, it looks ahead to the next stage of work.