Automatic detection of an early stage of basal stem rot disease infection using VGG16 and mask R-CNN

Abstract

Oil palm is the world’s most important oil crop, accounting for roughly 40% of all traded vegetable oil. Basal Stem Rot (BSR) has posed a significant concern to the oil palm industry, particularly in Southeast Asia, as it has the potential to cause substantial economic losses. Laboratory-based methods are reliable for early BSR detection. However, they are costly and destructive. Other methodologies used a semi-automatic approach which requires human intervention. Therefore, this paper presents an automatic detection of BSR using hyperspectral data and a deep learning approach, which includes a Mask R-CNN for image segmentation and a VGG16 as a classifier. The Mask R-CNN was trained using Set B images, and the images in Set A were masked using the mask produced by the Mask R-CNN. The VGG16 was trained with the masked images (Set A). This fully automatic approach demonstrated high model performance with 85.46% accuracy, 86.74% F1 score, 95.02% recall, and a classification time of 0.08s/image. The findings of this research have the potential to significantly benefit the oil palm industry by automatically detecting BSR at an early stage, thus allowing for the prevention of disease spread. It can also help solve the problem of labor shortage.