Performing classification processes in the food industry withdeep learning architectures

With the developments in the field of machine vision, the classification of many products in the industry is done using machine vision techniques. As a result of rapid developments in the field of deep learning, many problems that can and cannot be done with image processing have become possible with deep learning. Instead of image processing systems, which are long and difficult to integrate, deep learning-based models that can be easily adapted and keep up with the industry have been developed. In this thesis; CNN (Convolutional Neural Networks) models have been actively used in order to automatically perform classification processes of products in the food industry with deep learning techniques. The almond problem was chosen as an example problem; It is aimed to classify whole almonds, cracked almonds and almond shells with high accuracy. Thanks to the developed experimental setup, a total of 380 images were taken for each class of whole almonds, cracked almonds and almond shells. 70%, 15% and 15% rates were used for training, validation and test data sets, respectively. Alexnet, InceptionV3, MobilenetV3 and VGG16 models were used to train and test these datasets. As a result of the training, the highest test accuracy was the InceptionV3 model with 99.41%. In this way, faulty (unwanted) product detection in the food industry was made with high accuracy and an exemplary study was presented for classification processes.