Multi-spectral Image Fusion Method for Identifying Similar-colored Tomato Organs

Abstract

Aiming at robotic cultivation for greenhouse tomato, the multi-spectral image fusion method was researched, so as to enhance the image brightness difference between tomato's similar-colored organ, such as stem, leaf and green fruit, and simplify visual identification on them. According to the 300nm-1000nm spectral characteristics of the three kinds of organs, the Lasso penalty function was adopted to obtain the parameters of the functional logistic regression model for classifying the organ's spectral data, and the wavelengths of 450nm, 600nm and 900nm with the non-zero coefficients were selected as the optimal imaging wavelengths. A weight-fusion model on the multi-spectral images was proposed, and the fusion coefficients was estimated, through solve the problem to maximize the target-background difference and minimize the background-background difference based on NSGA-II algorithm. As result showed, the fusion results for recognizing the three organs all could enhance the brightness difference with the other background organs, and the ratios of image gray value SAD between the target-background and background-background organs in the fused images for fruit, leaf and stem respectively were 7.89, 13.56 and 2.06 times of the ratios in original images. The research was supposed to improve the identification on similar-colored plant organs under agricultural condition.