Atit Luksida and Prompong Sugunnasil
Published in Data Science and Engineering (DSE) Record 2025 Vol. 6 No. 1 pp. 517-530
Abstract
The current challenge facing factories in Thailand is the transition to Industry 4.0. The process of appearance inspection has been transformed from human inspection to a computer-assisted tool. The objective of this process is to improve the accuracy of the inspection by removing human judgment. In this study, we propose a convolution neural network (CNN) to detect the defect of electronic enclosure. Then, we compare the proposed method with several other techniques, including SVM and KNN. The testing dataset comprises 1,190 im-ages captured from a camera oriented in a consistent direction. These images were divided into four balanced classes to mitigate any issues related to class imbalance during model training. Although SVM demonstrated superior accu-racy, the substantial time required for training makes it impractical for real-world applications where time efficiency is crucial. In contrast, despite having slightly lower accuracy, CNN showed a beneficial balance between perfor-mance and computational efficiency, making it a more pragmatic choice in many real-world scenarios. KNN, although faster than SVM, had the lowest performance in our tests.