Identification of Anthracnose in Chillies using Deep Learning on Embedded Platforms

Chilli is among the most commonly used spices globally and is an integral part of many cuisines. Many countries like Mexico, India, China, and Korea are known for growing and consuming chillies. Amongst all, India is the largest producer of chillies worldwide. When cultivated on a large scale, these crops are highly susceptible to fungal, pests, weeds, bacterial, viral and pathogen attacks that substantially hinder production. Among these plant attacks, the most common is Chilli anthracnose, caused by the Colletotrichum fungus, which affects the leaves and the fruit of the chilli plant, causing a devastating loss to the farmers. Our paper proposes a solution based on Deep Neural Network (DNN) using transfer learning to classify disease-affected Anthracnose chillies from Healthy chillies. This study has developed a dataset by collecting the chilli samples from the University of Agricultural Sciences, Dharwad and chilli farms in Kusugal, outskirts of Hubli. The dataset consists of 4 classes with two types of chilli; red and green. Each coloured chilli has two stages; the healthy stage and the Anthracnose diseased stage. Here, different pre-trained DNN architectures and transfer learning methods are used to train the model on our dataset. Finally, the results are compared based on accuracy and model size for all architectures trained on the proposed dataset. And choose the architecture with the smallest model size and high accuracy for embedding in an edge device.