Deep Learning Image-Based Prediction

The problem addressed in this project is the lack of an efficient and automated system for the detection of Cassava Bacteria Blight (CBB) and Cassava Mosaic Disease (CMD) in cassava plants. However, conventional methods of disease diagnosis rely on visual inspection by trained experts which is subjective, time-consuming, costly, and limited by the availability of human resources. Moreover, visual diagnosis can be challenging due to the variability of symptoms, the influence of environmental factors, and the occurrence of mixed infections.

The deep learning model was based on a ResNet50 convolutional neural network (CNN) with multiple layers of filters, activation functions, pooling, dropout, and fully connected layers. The hyperparameters and optimization methods used to enhance the model’s performance and generalization were also discussed. The model was tested on a set of images and compared with existing models using different metrics. The prediction module was implemented on a web application that can be used by farmers and extension workers to diagnose and control cassava diseases in the field.