Research On The Key Technology Of Machine Vision In Agriculture Internet Of Things

China is an agricultural country and agricultural production takes an important position in China's economic production. With the further development of agriculture, the development of intelligent-agriculture has become the hot point. The development of intelligent-agricultural presents higher requirements for the application of computer, especially in the image recognition of agricultural environment. The existing relevant technology can t meet the growing demand. In this thesis a research on machine vision technology of the agriculture Internet of Things has been made. The key content of the research are the two important aspects of the image processing technology. One is image denoising and the other one is target recognition. The major works of this thesis are as follows:(1) The present development situation and the relevant background of Agriculture Internet of things has been discussed. By make analysis to the inadequacies of the present development situation of Agriculture Internet of things, we point out that the machine vision technology should be bring into the Agriculture Internet of things. After that the present development situation and research situation of machine vision technology was summed up and the inadequacies of the present research situation was pointed out.(2) The preprocessing method of image denoising was researched. A preprocessing method based on Bayesian decision was proposed. First, we established distinguish model based on Bayesian decision and then the image histogram was used to obtain the parameters needed for the Bayesian decision. To classify the pixels into two classificationsPPnoise pixels and non-noise pixels, the Bayesian decision was used twice. Then three kinds of traditional denoising algorithms was used to make the image denoising test. The experimental results show that the denoising effects was improved effectively by the preprocessing methods. It can improve the blurring of image in the traditional image denoising algorithms.(3) An improved anisotropic diffusion filtering algorithm was proposed. The Multi-direction Median Filtering Method which can strongly retain the edges and details was used to make diffusion in multiple directions. The local variance and image gradient was used to improve the diffusion coefficient. To enhance the robustness of the algorithm, the correct diffusion coefficient by several times of iteration was used. It can retain the edge details at the same time when filtering. 4 parameters( such as peak signal to noise ratio, mean square error, structural similarity and image number) was taken as indicators to make a compare. The simulation results show that the proposed algorithm is better than the traditional anisotropic diffusion method.(4) A research on crop target recognition in agricultural environment was researched and a method of recognition of ripe tomato was proposed. The hue component in HSI color model was used. The target image was transformed and the image was segment with maximum variance automatic threshold method. To promote the computational efficiency the smallest outer rectangle method was used to marked effective area. At last, the Hough transformation was used to recognize the contour. The background of the immature tomato and the branches and leaves was picked out automatically. The recognition effect of tomato which is blocked and is great and it can satisfy the demands of the actual agricultural production.(5) The method of tomato identification has been optimized. The results obtained by Hough transformation was modified with the least square fitting method to achieve the goal of improving the accuracy of recognition. After the contour was obtained the Hough transformation and least square fitting method was used to deal the results. The least square method was used to correct the result of Hough transformation. The experimental results show that the recognition rate of tomato was improved with the algorithm used.The two aspects of major works in this thesis were image denoising and target recognition. The results of research show that the effect of image denoising was improved and the target recognition of tomato was achieved. It sets the foundation for the following research.