DETECTION OF APPLE MARSSONINA BLOTCH DISEASE USING PARTICLE SWARM OPTIMIZATION

摘要

Apple Marssonina blotch (AMB) is a devastating disease that is predominantly found in Asian countries, such as Japan, India, and South Korea. The disease has been known to cause huge economic losses in the regions where it has been found. AMB causes early defoliation, which ultimately leads to low quality and quantity of harvested apples. In this work, spectroscopic measurements were collected and analyzed for two datasets from 2014 and 2015. A stochastic algorithm called particle swarm optimization (PSO) was used to find optimal features for classification. A total of ten spectral features were found by the algorithm by selecting pairs of bands that resulted in the highest discrimination between every two classes. A support vector machine classifier resulted in 100% classification accuracy for both healthy and diseased samples. Abundance estimation and spectral unmixing analyses of early-stage AMB (ambE) samples were also conducted using PSO to extract symptomatic and asymptomatic endmembers. Results showed reasonable separation between healthy, seemingly healthy, and symptomatic classes. Quantitative analysis, using varying degrees of infection of ambE samples, was performed by applying a combination of partial least squares and stepwise multiple linear regression models, and coefficients of determination (R-2) of 0.76 and 0.71 were achieved for the calibration and validation datasets, respectively. The results demonstrate the potential of using spectroscopic technology as a non-invasive method for early detection of AMB disease.