Automatic Rice Crop Extraction using Edge based Color Features and Color Indices

Crop segmentation is an essential stage of pre-processing for many automatic plant disease detection systems. Crop extraction process separates the vegetation from the background of real-time images taken from the field. The current work presents a color transformation-based region growing approach to segregate rice plant parts from common field background objects such as weeds, soil, shadow, and sky. The system is divided into three components performing weed identification and elimination, background separation, and restoration of misclassified foreground. The system is tested using 2223 rice plant RGB images. The method is designed to address two major concerns of real-time paddy plat processing; illumination variance of input images and the trade-off between soil and diseased plant part belonging to similar color families. The performance of the system is evaluated using SSIM, Jaccard Index, BF score and compared with other state-of-the-art color indices.