The use of deep learning in classification of agricultural products: Siirt pistachio case study

This thesis aims to investigate the potential of deep learning techniques in the classification of agricultural products and to develop an applied model. The agricultural sector generally relies on manual methods for quality control and classification processes, which makes these processes both time-consuming and error-prone. With the advancing technology, artificial intelligence-based approaches to automatically classify agricultural products have the potential to offer various advantages to both producers and consumers. In this study, a deep learning-based classification model is developed on the example of Siirt pistachio. In the first stage of the study, the quality criteria of Siirt pistachio were determined and a large dataset was created according to these criteria. The dataset was diversified to cover different quality classes (e.g., large, medium, low quality, deformed or discolored pistachios). The images were acquired at high resolution using advanced visual data acquisition techniques and data augmentation methods were applied to enrich the dataset. Convolutional neural networks (CNN) were chosen as the deep learning model and various architectural modifications were made to improve the performance of the model. During the training and testing of the model, hyperparameter optimization was performed and techniques such as dropout, learning rate adjustment and early stopping were applied to increase the accuracy rate. The performance of the model was evaluated using accuracy, precision, recall and F1 score metric. The results show that the developed deep learning model works successfully in the classification of Siirt pistachio with an accuracy rate of %99. The results of the model provide significant advantages in terms of both time and accuracy when compared to traditional manual classification methods. The potential for applicability to other products in the agricultural sector was also evaluated. This study is considered as an important step towards optimizing quality control processes in the agricultural sector, providing efficiency to producers and increasing consumer satisfaction. For future studies, it is recommended to create larger data sets, test different deep learning architectures and test the model in real-time applications.