Brown Planthopper Damage Detection using Remote Sensing and Machine Learning

Abstract

Every year paddy cultivators lose a significant amount of crop yield due to diseases and pests. Brown Planthopper (BPH) is one of the most common diseases that affect paddy cultivation. Sri Lankan government is struggling to make appropriate estimations regarding Brown Planthopper prevalence due to the absence of accurate and timely data. To solve this issue, a machine learning approach is proposed based on optical and synthetic aperture radar remote sensing data in this study. However, there is no previous effort for detecting Brown Planthopper attacks using machine learning and satellite remote sensing data under field conditions. This study consists of two phases. A time series classification based on SAR imagery is implemented to identify cultivated paddy fields in the first phase. Ratio and standard difference indices derived from optical satellite images are used in the second phase to identify regions affected by BPH attacks in paddy fields. Convolution neural network that is used in the first phase reports an accuracy of 96.20% for identifying cultivated paddy regions. A Support Vector Machine is used to detect areas damaged by BPH attacks in the second phase. The Combined approach of the first and the second phases shows promising results with an accuracy of 96.31% for detecting Brown Planthopper attacks.