Enhanced Deep Learning Approach for Potato Leaf Disease Detection

Potato is one of the most versatile staple crops which has the highest contribution of around 25.5 % of total crop production across the world. As agricultural technology advances and diseases of plants are diagnosed using artificial intelligence, it is critical to make research relevant to the advancement of sustainable agriculture. Potato plant farming faces challenges from diseases. Numerous leaf diseases including late blight and early blight, have remarkably impacted potato production rate. Diagnosis of such diseases manually is laborious and timeconsuming. In this paper, potato diseases are classified by applying Deep Neural Network architectures like VGG16, ResNet50, DensenNet121 and InceptionV3 on the potato dataset. Classification accuracy is evaluated with the augmentation method and without the augmentation method on training and test datasets. A comparative analysis of these four architectures is conducted to understand the performance issues. The empirical analysis revealed that all the models suffered from overfitting. To overcome this issue, an ensemble model is proposed by integrating ResNet50 with an autoencoder. This integrated model introduced a novel architecture that significantly enhanced performance. The proposed model outperformed state-of-the-art architectures on both training as well as test datasets.