Abstract:
Abstract: In order to solve the high workload and low efficiency problems while measuring the pigment content variation of citrus red mite infested leaves using the traditional physical and chemical methods, a novel pigment content measurement method for citrus red mite infested leaf using the hyper-spectral imaging technology was studied in this paper. In the research, 400 healthy leaves and 400 sick leaves were included as the test samples in which 350 healthy leaves and 350 sick leaves were utilized for model establishment and the other 50 leaves of each type were used for a model test. Each leaf's original spectrum and its first order deviation in its particular healthy and sick area were acquired to investigate the characteristic spectrum bands which could mostly reflect the variation of leaf pigment content. The correlation between characteristic spectrum band ratios and pigment content was analyzed. An univariate linear regression method was applied to analyze the pigment content prediction effect using the common vegetation indexes. A leaf pigment content prediction model was established, using the stepwise regression method, and the model's prediction ability was tested using the F test. Experimental results indicated that it is not satisfactory using the common vegetation indexes to predict leaf pigment content since they are not specially selected for citrus trees. The selected three characteristic spectrum band ratios of 667/522, 667/647, and 522/647 nm, each of which has a high correlation with a leaf's three types of pigment content, were applied in the stepwise regression method to establish pigment content prediction models. Two out of three of the characteristic spectrum band ratios of 667/522 and 667/647 nm, which gave the best performance, were used as independent values for model establishment. The F test results indicated that the established models could preferably predict both healthy and sick leaves chlorophyll a, chlorophyll b, and carotenoid content. The selected characteristic bands, as well as the established prediction models, could be used as the foundation to further study the citrus red mite infestation fast detection methods and techniques.