| Soybean is an important food and oil-bearing crop, it accounts for a great proportion in the people's lives. In our country, China is now the bigger producer of soybean and imported soybean country. The yield and quality of soybean influence the life of people in different stratum directly. and in order to ensure the yield, we must ensure the correctness of nitrogen. Symptoms of Nitrogen deficiency or excess will be shown in soybean plants'external morphology, many literatures have reported that soybean plants under the Nitrogen deficiency or excess, there are some bronze disease spots on leaves, leaves will turn yellow and dry, stems of plant are thin etc. we determine nitrogen fertilizer by keeping track of soybean leaves in different periods. In this way, there will be a correct subjective direction that the amount of nitrogen applied.At present, it is mainly through the naked eyes to diagnose disease and judge the nitrogen contents. The same is true of plant pathology in which subjective judgment plays a dominant role. As a result of personal levels and experience, there tend to be different conclusions based on the same sample. Considering its high subjectivity, lack of exactitude, as well as the complicated and time-consuming procedure, this method has no longer meet the need of modern agriculture. Thus, developing a fast and accurate method of detecting the nitrogen contents in leaves is of great practical significance.In this paper adopts sand culture method to cultivate the soybean plants that nitrogen content is 0%, 50%, 100% and 150% respectively, and we got large number of samples in different stages of growth and different nitrogen content. Soybean leaves sampling by the scanner to scan the leaves. The growth of soybean can be divided into V stage and R stage. V is the vegetative growth stage, V2 means stage of two leaves, V3 means stage of three leaves; R is the reproductive stage, R1 is beginning bloom, R2 is full bloom, R3 is the beginning of pod, R4 is full pod stages, R5 is initial seed filling stage, R6 seed filling stage. In this study, we collected V2, V3 and R1~R4 stages six different of the leaves with different nitrogen content, to establish sample library of leaves for feature extraction.In this thesis, the pretreatment methods, such as image grayness, gray equilibrium, threshold division, image smoothness are introduced in detail, and takes the method of modified maximum classes square error separate the leaf from the background. In order to reduce the disturbance produced by the illumination change, after comparison, a method was proposed, selecting B component under the RGB color model to conduct the separation of leaves. Through the features of color, texture, shape combined with the analysis of the characteristics, the symptoms of leaves are described effectively. There is 27 color features, extracted mean value, deviation, third-order moment of RGB, HSV and Lab color components; 8 texture features; include energy mean, energy deviation, correlation mean, correlation deviation, homogeneity mean, homogeneity deviation, difference mean, difference deviation; 5 shape features, include the average length of ellipsoid axial ratio, compaction mean, the average inside diameter ratio, degree of deformity mean, circular-degree mean, up to 40 characteristics.The paper established a three-layer BP neural network. The extracted parameters were then standardized and the principal component of the parameters are analyzed which remarkably reduced dimension of the input data. Identify the nitrogen contents in leaves of six different times, picked 360 leaves of every period as the input, a test based on neural networks of 120 leaves of every period, shows that average recognition precision is 93.6%, which will meet the actual diagnosis demand. |
PDF Full Text Request