关键词: 冠层叶面积, 光场重聚焦, 大豆植株, 图像处理, 回归模型

Abstract: Soybean upper, middle and lower canopy leaf area distribution is an important basis for soybean plant type status assessment and yield prediction. Traditionally, the field slicing method was used to measure the upper, middle and lower canopy leaf area, which was cumbersome and could cause certain harm to the leaves. In order to solve this problem, the refocusing technology of light field camera was introduced to obtain the refocusing images focused on the upper, middle and lower leaves respectively. The focal plane blades were extracted by image processing technology and the leaves of the defocusing planes were removed to get the upper, middle, lower projection area respectively. In this study, 103 pots of Hongqiu variety soybean at flowering stage were selected as calibration set, and the calibration coefficients of each canopy leaf were calculated to obtain the revised projected area of each canopy leaf according to the calibration of light field camera. A regression model was established for each canopy projection area and the true leaf area of soybean plants, and 20 pots were selected as prediction sets to verify each regression model. The study found that the coefficient of determination of upper leaf area model was 0.945, the maximum error of prediction set was 4.48%, and the root mean square error was 4.376. The determination coefficient of the middle leaf area model was 0.796, the maximum error of the prediction set was 13.62%, and the root mean square error was 7.273. The lower leaf area model had a coefficient of determination of 0.914, the maximum error of the prediction set was 8.63%, and the root mean square error was 1.529. The correlation of the upper and lower leaf area measurements was high, and the correlation of the mid-leaf area model was lower due to the occlusion of the upper leaf.

Key words: canopy leaf area, light field refocusing, soybean, image processing, regression model