Vision Based Solution Towards Resistance Value Determination

Identifying resistor values through color bands is essential in electronics production; however, conventional techniques encounter challenges regarding precision and scalability, particularly in practical environments. The proposed work aims to provide a scalable and resilient deep learning solution for automated detection of resistor values, tackling issues related to different lighting, resistor angles, and image distortions. The proposed work utilized MobileNetV2 framework, which has been pre-trained on ImageNet, and refined it with a tailored dataset to improve efficiency in dynamic and industrial environments. A custom dataset was built consisiting of images of resistors captured in various lighting conditions and angles to ensure the model’s capacity to generalize. The optimized MobileNetV2 model shows exceptional performance, attaining an accuracy of $\mathbf{9 6 \%}$ in controlled settings and $\mathbf{9 2 \%}$ in real-world scenarios. The results are demonstrated with recall and F1 score, emphasizing its reliability in identifying resistor values in various situations. The lightweight design of MobileNetV2 enables quick inference, making it suitable for real-time applications, even in environments with limited resources. The proposed method greatly surpasses conventional techniques and CNN-based models, providing a scalable, efficient, and reliable solution for resistor detection, with possibilities for incorporation into automated industrial systems to improve production efficiency and quality management.