Combining Weather Classification and Mask RCNN for Accurate Wheat Rust Disease Prediction
Deepak Kumar, Vinay Kukreja, Bhawna Goyal, Shanmugasundram Hariharan, Aditya Verma
Abstract
Wheat rust diseases cause a huge amount of loss and decrease the wheat yield quality. According to the national agriculture institute, a total of 6.75% of wheat grain quality is decreased due to wheat stem, stripe, and leaf rust diseases. Thus, the identification of rust diseases is important for farmers. The rust diseases have been identified by either evaluator or traditional image processing techniques and this takes a lot of misidentification errors. To reduce the misidentification errors, a Mask RCNN M2 approach is proposed. The Mask RCNN2 approach is validated on the collected secondary source dataset with three numbers of weather classes such as rainy, cold, and sunny classes. The Mask RCNN M2 is a combined approach of Mask RCNN and CNN for rust diseases classification with a different number of weathers. The weather classification defines the spreading of rust diseases with different number of weather. If the rust disease along with weather is classified, then the spreading of rust diseases in different weather is controlled in an effective manner. A total number of 3600 wheat rust masks with a size of 36$\ast$36 pixels have been generated. The generated masks have been used for weather classification. The weather classification helps find the spread of rust diseases in different weathers. Several epochs were used to validate the Mask RCNN2 approach. During epochs tuning, Mask RCNN has high validation loss (0.25) and CNN achieves high testing accuracy (98.6%) for weather classification. Thus, the current study facilitated the control of the identification of wheat rust diseases in different weathers.