Plant Leaf Disease Identification by Deep Convolutional Autoencoder as a Feature Extraction Approach

A healthy plant is needed to provide good quality of food which is important to daily life. However, plant diseases have a major impact on agricultural production and economy. Conventional approach is mainly based on hand labor experiences of farmers leading to that it may have adverse consequences in preventing disease transmission. Various techniques have been applied to identify the disease in plant in order to provide solutions to the farmers as well as send them a warning signal. This paper proposes a method for plant disease identification through a set of collection of leaf images. This method is developed based on a deep convolutional autoencoder approach. Original images are firstly reconstructed by the autoencoder and then by using only encoder part, the features of these images are extracted. The output features are fed into Support Vector Machine for classification. A small set of Plant Village dataset is considered to verify the performance of the reconstructed performance by the convolutional autoencoder. Then, for the classification stage, simulation results obtained have shown that the difference of autoencoder architecture and kernels of classifier give different average accuracy. Thus, the highest accuracy reached 98.8% for plant disease classification.