Determining the quality levels and species of trout, sea bream, and bass fish by machine learning

Depending on the processing of the concept of healthy nutrition with the results of scientific research, the importance of fish meat consumption for public health increases day by day. The essential effects of unsaturated fatty acids found in fish in human metabolism make it more functional than other foods. However, the complex structure of the functional compounds in question turns fish meat into a perishable food. Accordingly, the quality parameters of fish meat must be determined very well and quickly. Sensory, chemical and microbiological analyzes used to determine fish quality are specific analyzes that require time, equipment and trained personnel, and the results are delayed. Machine learning, on the other hand, is artificial intelligence consisting of various algorithms that can process information inputs and turn them into useful information for the user and use them to make predictions. With the developing technology, machine learning has initiated significant changes in many industries and has enabled the elimination of important problems with its high performance. The biggest advantage of machine learning is to process large amounts of data quickly and have a time-saving checkpoint. In determining food quality, machine learning can be an important method that can be applied without contacting and destroying food. In this thesis, it was aimed to determine the quality parameters of different types of fish by using deep learning method. Different types for this process; Trout, sea bass and sea bream were stored in ice for 7 days (3±1°C) and their pictures were recorded on each storage day. A total of 21,386 images were obtained during storage for 3 different fish species. These images were subjected to quality classification by machine learning. In the study, an algorithm known as SVM (support vector machine), which is a support vector machine, is used to classify images using CNN (Convolutional neural networks) features obtained from images. CNN's first layer specified the input dimensions of the images. Accurate classification rates ranged from 90.6% to 100%, and it was observed that the expected increase in accuracy was achieved with the increase in data. Average correct classification rates for trout, sea bream and sea bass were above 80%. The results of the study showed that the fresh-stale distinction of fish can be determined with high accuracy with machine learning and machine learning can be used as a fast quality determination method.