Apple Varieties Classification using Light Weight CNN Model

Fruit classification is a vital task in many industrial contexts. The automated fruit classification system is becoming more and more important in the food-processing sector. In this paper, we offer a method for sorting apples using deep learning technology. Deep learning models can achieve cutting-edge accuracy, sometimes even outperforming human ability. Using convolutional neural network (CNN) architecture, the proposed model will be able to identify 14 different varieties of apples. For feature extraction, five layers of two residual blocks are used. Artificial neural network (ANN) based neurons used for the classification of apple varieties in fully connected layers. The CNN layers used in residuals are a combination of convolutional layer, non-linearity layer, pooling layer, and fully connected layer. When compared to ResNet50, VGG16, MobileNet, and EfficientNetB0 state-of-the-art architectures, the proposed approach is able to obtain improved training accuracy of 99.95%, validation accuracy of 99.02%, and testing accuracy of 99.59% with adam optimizer.