Fruit-Net: Fruits Recognition System Using Convolutional Neural Network

AbstractFor many industrial applications, classifying fruits is an essential process. A supermarket cashier can use a fruit classification system to distinguish between different types of fruit and their prices. Additionally, it can be used to determine whether a particular fruit species satisfies a person’s nutritional needs. In this chapter, we propose a framework for fruit classification using deep learning techniques. More specifically, the framework is a comparison of two different deep learning architectures. The first is a 6-layer light model proposed for convolutional neural networks, and the second is a carefully tuned deep learning model for group-16 visual geometry. The proposed approach is tested using one publicly accessible color-image dataset. The images of fruit that were utilized for training came from our own photos, Google photos, and the data that ImageNet 2012 gave. This database contained 1.2 million images and 1,000 categories. The 1,200 fruit images that had been divided into six groups had been assessed and categorized. The average classification performance was 0.9688 out of a possible range of 0.8456 to 1.0 depending on the fruit, and each photo took about 0.25 s to classify. With only a few errors, the CNN algorithm was able to successfully classify the fruit photographs into the six categories. On the dataset, the CNN, VGG16, and Inception V3 models each achieved classification accuracy results of 96.88%, 72%, and 71.66% respectively.KeywordsFruit recognitionConvolutional neural networkClassificationVGG16Inception V3