Image Identification Research Of Cotton Leaf Characters Based On Machine Vision

Leaf is the important apparatus of plant for photosynthesis and transpiration, and many information of leaf reflect the growth status of plant. This information is very important to control the growth of the plant. This research regarded the cotton leaf as the study object, on the aid of nation "863 plan" (200060110Z2053) and nation science foundation (30360047), based on the machine vision processing theory and technology, combined routine test data in agriculture, studied the methods in attaining images, reducing yawp in image, division image, and building models in computing geometry and form characters and nitrogen quantity of leaf. It offered a lot of important information for plant, and avoided of complex and slow plant experiment.The primary researches in this study included:1. Studying the shooting type of the leaf. This research established the rule of shoot in black box on the results of experimentation, so the problem of computing nitrogen is avoided by great difference of color in different circumstance.2. Studying the method of reducing yawp of image. This thesis compared the theory and application result in cotton leaf digital image of average and middle of box, and made the conclusion that using middle box method can remove the spot in the image, and not change the color of leaf.3. Image division is threshold method in this research, studying the theory and method of threshold. Because of the histogram of the image, comparing of the division effect and time of division based on static RGB threshold, dynamic G threshold, static rgb threshold, static rgb threshold combined dynamic I threshold, and dynamic G threshold combined static H threshold this five methods, found the division suited for this research.4. Studying the technology and method of computing leaf geometry and form characters based on image character and image maker. This thesis designed the structure and array to select the valid data in image pixels, used the linked code and pixel area methods to compute the area, perimeter, height, width, diameter, rectangleness, and roundness, etc.5. Studying the method of using artificial nerve network to build model between color character and nitrogen quantity. After compare the train effect and forecast effect of line network, BP network, and RBF network with six pieces of input vectors in a great of data, this thesis select the best result. To improve the precision of result, this research studied the algorithmic theory and parameter.The primary results of this research include:1. Getting the image division based on dynamic G threshold combined static H threshold, the result and time is very good.2. Computing leaf girth, area, height, width, diameter, proportion of height and width, roundness, rectangleness and nitrogen quantity indexes, and the precision of area, height, diameter, width and proportion of height and width is very high. It used RBF network with (B, H, G-R, G/R) as input vector, building the math model between color characters and nitrogen quantity, using this model to test 30 pieces of images, the correlation coefficient between result of network and result of agriculture routine test is very marked level (r= 0.8674**).3. Implementation the cotton leaf digital image processing system, can get the best division effect for cotton leaf image, and compute nine agriculture characters.