Detection of rice stem borer infestation based on hyperspectral radiometry via machine learning algorithms

Stem borer is one of the major insect pests of rice in Asia whereby heavy infestations can affected up to 80% of rice grain yield loss under field conditions. The larvae bores and feed upon tillers that resulted in 'dead heart' or dying at the central tiller during vegetative stage and infestation at reproductive stage caused panicles to become unfilled or 'white head'. Hyperspectral radiometry stores continuous spectrum of narrow and contiguous wavelength bands which allows to distinguish subtle changes in plants reflectance properties. Paired with machine or deep learning algorithms, these features allow the detection, identification and estimation of various crop stresses through analysis of their spectral reflectance properties. The detection of stem borer infestation traditionally depends on laborious manual monitoring and early infestations are difficult to be detected visually. Therefore, to improve detection accuracy and reduce human error, this study aims to determine the changes in spectral reflectance between healthy and infested rice plants. Experiment was carried out with infestation by yellow stem borer larvae, Scripophaga incertulas (Walker) on MR315 rice variety in Greenhouse Complex, UPM (Mean T:29 deg C, RH: 81%). The spectral reflectance data from 8 healthy and 24 infested hills were recorded at 5, 10, 20, and 30 days after infestation (DAI) using portable visible shortwave near infrared (VSNIR) spectrometer. The spectral reflectance samples were pre-processed and analyzed using Orange3-3.35 Data Mining software. Several learners methods were employed to classify between healthy and infested samples, in addition, statistical methods were utilized for validation of classification results. The result shown spectral reflectance of infested tillers gave significant decrement in near-infrared (NIR) region compared with healthy plants and Random Forest (RF) scored the highest classification accuracy (0.958) as early as 5 DAI.