Disease Detection of Bangladeshi Crops Using Image Processing and Deep Learning - A Comparative Analysis
Nafisha Binte Moin, Nabila Islam, Shamima Sultana, Lubaba Alam Chhoa, S. M. Ruhul Kabir Howlader, Shamim Ripon
Abstract
Crops diseases can have many adverse effects on the economy and management of food resources in a developing country like Bangladesh. Bangladesh is an agricultural-centric country where Rice, Potato, Corn/Maize, and Wheat are a few of the major crops. Crop diseases can cause low yield of food which can cause harm to the people of a country. Proper disease detection is essential but there remains a challenge in identification of crop diseases and its proper treatment. With the advancements of image processing and deep learning, crops diseases can be detected in a matter of seconds. In this study, several experimentations have been done for both diseased and healthy crops images using six widely popular CNN models namely Xception, VGG16, ResNet152V2, InceptionResNetV2, DenseNet201, and MobileNetV2 for four separate major crops of Bangladesh. An accuracy of 95.52% has been achieved for Corn and 98.55% accuracy is secured for Potato by applying Densenet201, while 64.30% is achieved for Rice. On the other hand by applying MobileNetV2 an accuracy of 98.28% is obtained for Wheat. The experiments are conducted for both colored and grayscale images. However, notable improvement is not observed for images that are converted to grayscale. Significant improvement in disease detection has been observed for most of the corps except for Rice. A further improvement has been achieved by applying SAM.