Custom Residual Networks and Transfer Learning: Tea Leaf Disease Detection

In Bangladesh, the tea industry holds significant importance both economically and culturally. Tea leaf diseases represent a major challenge for the tea industry in Bangladesh. We developed and compared image detection models for detecting tea leaf diseases, specifically custom Residual Networks (CRN) and ResNet50 models. The paper aims to evaluate the performance of these models in accurately identifying tea leaf diseases. The pre-processing and data augmentation techniques increase the model’s ability to detect diseases effectively. The accuracy and performance in disease detection by training and testing both the custom ResNet and ResNet50 models on the tea leaf image dataset. The proposed CRN architecture proved to be more precise in testing (51%) than the ResNet50 architecture, verifying its effectiveness. The accuracy achieved is acceptable for the suggested model. A confusion matrix was also used to display the performance of the model and confirm the accuracy achieved.