基于随机森林算法的整株水培生菜缺素诊断

Abstract

The phenotypic information of lettuce leaves can well reflect its health. In order to diagnose the nutrient deficiency types of hydroponic lettuce accurately, non-destructively and quickly in the mid-growth stage, a method for diagnosis of whole lettuce based on random forest algorithm (RF) was proposed. The images of lettuce under four different conditions, K-deficiency, Ca-deficiency, N-deficiency and Normal, were collected and segmented by Extra-green algorithm. Then, features of color, texture and shape were extracted. A RF classification model for the hydroponic lettuce nutrient deficiency diagnosis was constructed and compared with support vector machine (SVM) and back propagation neural network (BP). RF had the best classification effect among the three methods. The overall classification accuracy was 86.32%, Kappa coefficient was 0.82, and it can provide a basis for the prevention and remedies of lettuce deficiency and the scientific management of nutrient solutions.