Smart Farming Solutions: An Integrated Approach to Agro-Product Quality Assessment using Thermal Imaging and Machine Learning

Abstract

Classification of Potatoes into Defective and non-defective is an important area of impact for agriculture based industries. In this paper, quality of agro-products are assessed to aid in the development of neural network and thermal imaging based automated computer vision system. For classification, object features are extracted from thermally generated images. The primary dataset of 480 thermally generated images is developed. Various colour and statistical based features are used as an input to the neural network. Two hidden layers based Artificial Neural Network (ANN) based models are used. Impact of each feature is determined by considering a single-input ANN. Multi-input ANN based models are deployed to assess the performance of each features in combination with other features. Entropy and Energy provides high classification accuracy. Variance is observed to be a poor feature for classification. Considering all the features, except variance, 98.4% of classification accuracy is achieved.