Classification of cable colors and verification of positions with artificial intelligence based real time image processing

In this thesis, an artificial intelligence-supported image processing system was developed to prevent faulty cable soldering problems encountered in electronic manufacturing processes. The system aims to provide real-time quality control by classifying whether cables of different colors soldered onto printed circuit boards (PCBs) are positioned correctly. Cable connections are categorized into two classes: OK (correct) and NOK (faulty). Widely used deep learning models such as VGG16, MobileNetV3, YOLO, and ResNet-50 were trained and evaluated on labeled image datasets containing various color combinations of cable connections. During training, different values of key hyperparameters—including epoch number, batch size, and learning rate—were tested. The performance of the models was compared using metrics such as accuracy, precision, and F1 score. According to the test results, the VGG16 model achieved the highest performance with 97.6% overall accuracy and 1.00 recall for the NOK class. The developed system operates in real time when integrated into the production line, reduces the need for manual quality control, and aims to prevent material losses caused by errors. This study serves as an important example of the effectiveness of artificial intelligence and image processing techniques in industrial applications and contributes to smart manufacturing systems.