Lite Transfer Learning Model for the Efficient Classification of Plant Leaf Diseases using Few Samples

It is important for farmers to investigate the growth and health of the crops in order to take necessary and timely actions regarding any disease or pests. An automated classification technique is necessary for the faster detection and prevention of diseases in plants. Convolutional neural networks based deep models give good performance in the plant disease classification tasks compared to the traditional feature extraction techniques. This advancement of deep learning is possible due to the availability of large annotated datasets. Lack of annotated samples may result in performance degradation. To tackle this problem, the researchers are proposing few-shot learning concepts. This technique focuses on identifying unseen classes during training with annotated data. This paper classifies plant diseases by considering few samples and lightweight transfer learning based deep learning models that could be effectively installed in the resource constraint devices. We propose a few-shot classification technique to effectively classify plant diseases using the plant village dataset. This dataset comprises of 38 classes of leaf diseases, having 54,303 leaf samples. Different approaches were considered for dividing the dataset into the source and target domains. Extensive experimentation is perfomed on the target dataset considering various sample sizes. MixNet-XL regarded as the base model for this work, and a customized circle loss is used to classify the samples effectively. This work considers four domain splits for different comparison purposes, and it is found that accuracy is improves by 1.438 % for split-1, 15.57% for split-2, 1.3% for split-3, and 2.02% for split-4 when considering proposed methodology. The model presented here is compared with several recent works using different evaluation criterias, and sample sizes, and satisfactory result is obtained in each case.