A Comparison of HOG-SVM and SIFT-SVM Techniques for Identifying Brown Planthoppers in Rice Fields

Abstract

Brown planthoppers (BPH) are insect pests that cause significant damage to rice crop yields throughout the Asia-Pacific region. Early identification of BPH forms has ramifications for forecasting potential outbreaks. To address this, we use Adaboost and Haar features to discover areas of interest in images of rice plants. We apply two separate techniques to identify the BPH in images: we compare a technique that utilizes HOG descriptors and another that utilizes SIFT feature descriptors. To each of these techniques, we apply a Support Vector Machine (SVM) to allow us to classify areas of interest in the images. Our approach achieves a weighted average classification rate of 95.38% for HOG and 96.38% for SIFT, improving upon state-of-the-art BPH detection methods and our findings lay the groundwork for other insect pest identification and detection efforts.