Research On Corn Variety Identification Based On Depth Image

As one of the world’s three major crops, corn plays an important role in the world’s sown area and production, which is one of necessary food crop and oil crop. China is a large agricultural country, which possesses many kinds of crops. Corn is not only one of the largest sown area of crops at present, but also the foundation of the entire agricultural production, which occupies a pivotal position in the national economy. Corn seed is the key to develop corn production, for the reason that the seed quality has a significant effect on the production. At present in China, the identification of seed breeds authenticity has a special effect, especially on researching a rapid identification method which can be applied to identify the authenticity of seed breeds in seed trade flows and planting. In the meanwhile, it is important to guide agricultural production, protect farmers’ interests and ensure national food security. In recent years, computer vision has been widely applied to the detection and identification of varieties of crop seeds. By using computer vision technology, we can identify automatically the different corn varieties.Due to irregular shapes, variable colors, different sizes of corn grains, poor consistency in the same kind of seeds, more characteristics of different seeds, the possibility of miscarriage justice and the identification of different corn varieties are very difficult. Based on digital image processing and pattern recognition technology, depth images of corn grains were studied and the effective characteristic parameters of corn grains were found. The main works are as follows:1. The depth images of corn varieties were obtained and the color space was chosen. The shooting conditions of the depth images of corn varieties and the space model of color were studied. At the same time, the paper designed shooting method and achieved a good result that was color feature of the depth images in corn varieties on the HSV color space model after experimental verification.2. The preprocess method of the depth images of the corn varieties were confirmed. In order to make depth information of corn grains prominent, and make it easy to extract feature, it’s necessary to preprocess the color depth of the generated images. Image graying, image enhancement, image segmentation and Harris algorithm were used to detect the edge of corn grains. The desired effect of pretreatment was achieved.3. The depth image features of corn varieties were extracted. A problem which was that directly graying the depth images existed certain errors was found through the study. This thesis proposed that combine the color feature with graying the depth images, extract the color features, gray features and the tip of the perimeter, area, circularity shape features.4. The method of the corn species identification was studied. According to the characteristics of the parameter selection, the histogram based on the characteristics of color and BP neural network based on the shape of the tip of corn grains recognition was built, and analyzed and compared the recognition results. By combining the two methods, all the feature parameters were put as the input of neural network and the corn varieties were identified. The experimental results showed that the accuracy of combination with the tip shape feature and color histogram was higher.