关键词: 生菜, 支持向量机, 机器学习, 热成像技术, 水分胁迫检测

Abstract: During the growth stage of lettuce, water stress detection on lettuce plants can effectively save water resources without affecting the quality. In this study, 140 lettuces 1901 and Yeluo were used as experimental subjects, and different irrigation amounts were applied during the growth stage of lettuce. Thermal imaging technology was used to obtain lettuce canopy temperature information, and the maximum value, minimum value, mean value, variance, standard deviation, entropy value, coefficient of variation and frequency values of different temperature widths based on the plant canopy temperature information were used as characteristic values. Support vector machine （SVM）, Random Forest （RF）, Whale Algorithm Optimized Support Vector Machine （WOA-SVM） and WOA-SVM based on principal component analysis were established to detect the water stress degree of lettuce, and the accuracy was compared. Experimental results show that RF has an accuracy of 94.76% and 92.37% for Yeluo and 1901. SVM has an accuracy of 91.64% and 85.35% for Yeluo and 1901. WOA-SVM has an accuracy of 98.77% and 94.76% for Yeluo and 1901. The PCA-WOA-SVM model has an accuracy of 98.94% and 95.71% for Yeluo and 1901. The PCA-WOA-SVM recognition accuracy is high and stable.

Key words: lettuce, support vector machine, machine learning, thermal imaging technique, water stress detection