Proximally sensed RGB images and colour indices for distinguishing rice blast and brown spot diseases by k-means clustering: Towards a mobile application solution

Abstract

Rice blast (RB) and Brown spot (BS) are economically important diseases in rice that cause greater yield losses annually. Both share the same host and produce quite similar lesions, which leads to confusion in identification by farmers. Proper identification is essential for better management of the diseases. Visual identification needs trained experts and the laboratory-based experiments using molecular techniques are costly and time-consuming even though they are accurate. This study investigated the differentiation of the lesions from these two diseases based on proximally sensed digital RGB images and derived colour indices. Digital images of lesions were acquired using a smartphone camera. Thirty-six colour indices were evaluated by k-means clustering to distinguish the diseases using three colour channel components; RGB, HSV, and La*b*. Briefly, the background of the images was masked to target the leaf spot lesion, and colour indices were derived as features from the centre region across the lesion, coinciding with the common identification practice of plant pathologists. The results revealed that 36 indices delineated both diseases with 84.3 % accuracy. However, it was also found that the accuracy was mostly governed by indices associated with the R, G and B profiles, excluding the others. G/R, NGRDI, (R + G + B)/R, VARI, (G + B)/R, R/G, Nor_r, G-R, Mean_A, and Logsig indices were identified to contribute more in distinguishing the diseases. Therefore, these RGB-based colour indices can be used to distinguish blast and brown spot diseases using the k-means algorithm. The results from this study present an alternative, and non-destructive, objective method for identifying RB and BS disease symptoms. Based on the findings, a mobile application, Blast O spot is developed to differentiate the diseases in fields.