ABSTRACT
Optimum crop mapping and monitoring solutions have been provided by combining temporal remote sensing images with machine learning approaches. Monitoring the spread and extent of crop disease is an important aspect of food security. Muzzafarnagar district in Uttar Pradesh, India, is rich in sugarcane cultivation. Sugarcane fields have varying germination processes which do not allow statistical-based machine learning algorithms to handle high levels of heterogeneity. An individual sample as a mean training method is used in this study for training the machine learning model which considers the effect of each pixel individually and not as a statistical measure. This study utilizes a temporal indices method to map the diseased sugarcane ratoon fields with a fuzzy kernel-based modified possibilistic c-means classifier. The non-uniform spread of disease makes it challenging to map the disease-affected fields from optical satellite data. The PlanetScope DOVE (3-m) sensor dataset is utilized for this study. This study shows that the healthy and diseased sugarcane ratoon fields can be mapped using the proposed approach with mean membership difference values of the order of 0.04 and 0.02 for healthy and diseased sugarcane ratoon fields, and it aims to utilize the temporal satellite data for unique crop type identification and mapping.
