Fruit and Vegetable Quality Detection Using SENet and EfficientNet

Food safety and management of the food supply chain heavily depend on fruit and vegetable quality detection. Generally, this function is undertaken by retailers or dealers. Computer vision systems can also assist humans, however, since humans can also have fatigue from such boring and repetitive tasks, their efficiency may decline and more importantly, mistakes may be made. Conventional Convolutional Neural Networks (CNNs) use complex strategies to deal with huge volumes of information and mostly discriminate between the images of fruits and vegetables describing them as fresh or rotten. In this work, we present a novel approach based on a hybrid model comprising of SENet and EfficientNetB0 allowing us to determine the class (fresh or rotten) of fruits and vegetables. The Fresh and Rotten Classification dataset consists of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{2 1, 0 0 0}$</tex> images of fruits and vegetables with different levels of freshness. The study used images of apples, bananas, cucumbers, okra, oranges, potatoes, and tomatoes. This dataset consists of 14 fresh and rotten image classes. When we analyze the results in the case of traditional CNN models, the results of the proposed model are much better, specially that large datasets can be handled and central processing units in other hardware used more efficiently.