An Automatic Rice Plant Disease Detection Model Built With Unstructured Data Using IMDT Tiling and CNN Cognitive Object Detection

Nowadays agriculture and processes are getting more intelligent mechanisms to improve the yield and reduce manual work. Smart agriculture provides numerous modern ideologies to farmers. But still, farmers face one important issue crop disease. Many researchers provide plenty of ways to recover and tackle the situation to come out of this problem. Therefore, they proceed with image processing to identify diseases from rice plant images. Farmers mainly face problems to take proper images for classification. Because of various reasons like various environmental factors, farmers ignorance, field size, capturing angle, device limitations, etc. are affecting the quality of the disease detection system, and these factors degrade overall performance. For this problem, introducing the Intelligent multi-dimensional tiling (IMDT) technique with an advanced convolution neural network with cognitive object detection (CNN-COD). IMDT technique developed with an intelligent expert system that adjusts input image size, capturing angles and other factors automatically. This advanced tiling technique supports to do the cropping and fluttering of input images for resizing. And CNN-COD model was used to calculate rice leaf width size and rescaled at the time of image segmentation with the Residual network (ResNet) model. Created dynamic tiled images are uniformly and scaled dimensional objects. These input values are used to train the CNN-COD rice plant disease, prediction model. Our proposed models were appraised on more than 4960 images which contain 8 various types of rice crop diseases. The experimental result portrayed out the CNN-COD model receives significant improvement in objection detection and image classification for the rice plant disease detection system. Mean average precision (MAP) values compared the CNN-COD model with the YOLOv4 model it got improved by 3.7% with the tiled input dataset.