Rice Nutrition Identification And Diagnosis Based On Machine Vision Technology

Recently, the digital diagnosis of plant nutrition is one of the hot topics of plant nutrition and agriculture remote sensing, so it is important to establish a simple, accurate and rapid diagnosis method of crop nutrition. In this study we propose two platforms to get the image of rice leaf and sheath.one is the flatbed scanner, the other is the portable scanner. The technology of machine vision is used to extract characteristics of rice leaf and sheath, and according to the relation between the characteristics and NPK nutrition status of rice to choice the specific characteristics set for modeling. Additionally, digital camera and measurement play the important role in modeling the recognition of nutrition pattern. The results are as follows:1. Selection of method for obtaining digital imagesThe flatbed scanner was selected to collect images of rice leaves and sheaths. Scanning was carried out in a closed environment and can avoid interference of the external environment. Scanning can accurately, fast acquire color, shape and texture information of rice leaf and sheath, and some subtle characteristics as leaf plot. Scanning provides a new method for digital diagnosis of rice nutrition. For verifying the applicability in the field we use portable scanner to obtain the image of rice leaf to diagnosis of rice nutritional status.2. Identification of NPK nutrient stress based on scanning of leaf and sheathRice leaf and sheath showed significant variations at different nutrition status. After analyzing the characteristics of the digital image, leaf and sheath color information, leaf and sheath shape information, edge of leaf information and spot area information played an important role in identification of the N, P and K nutrition. In addition, Fisher Discriminant Analysis was used to establish the identification rules for different nutrition status. The results showed the highest training accuracy of identification of NPK nutrition respectively were86.15%,87.69%,90.00%and89.23%in4growth periods, the validation accuracy respectively were83.08%,83.08%,89.23% and90.77%. Note that the old leaf can get a better result compared to the new one in the all growth stages.3. Identification of NPK nutrient levelsAccording to the mechanisms of N deficiency, we selected GL, LengthL, AreaL. GLT. GLS, and Lengths as the diagnosis characters to identify the N nutrition levels. The accuracies of identification respectively were94.00%,98.00%,96.00%and100.00%in4growth stages.According to the mechanisms of P deficiency, we selected Lightness, AreaL, lengthL/s, AP_ratio, RLS, LengthLS and LS2"3as the diagnosis characters to identify the P nutrition levels. The accuracies of identification respectively were94.00%,92.00%,98.00%and94.00%in4growth stages.According to the mechanisms of K deficiency, we selected GL, GLT, AreaLMVR and Spot as the diagnosis characters to identify the K nutrition levels. The accuracies of identification respectively were96.00%,90.00%,94.00%and94.00%in4growth stages.