Research on Automatic Focusing Technology for Knife Defect Detection Systems

In the knife defect detection system, accurately determining focal distance at high speed is crucial for obtaining clear images, necessitating further algorithm optimization. We introduce an innovative autofocus mechanism, ST-VGG, leveraging an optimized deep-learning model based on the VGG16 architecture. This model enhances convolutional layers, employs global average pooling to mitigate overfitting, and integrates the Inception architecture for multi-scale feature fusion without additional computational load. By using depthwise separable convolutions, ST-VGG effectively reduces parameter complexity, making it an ideal solution for mobile devices and embedded real-time detection. This innovative approach converts autofocus tasks into regression problems, enhancing prediction accuracy compared to leading lightweight architectures such as TVGG, TSwinT, and TViT, as well as classic models like LeNet and AlexNet. Additionally, the ST-VGG model demonstrates remarkable potential in knife defect detection systems by effectively addressing the speed bottleneck in the autofocus process. It achieves notable results on public datasets, reducing verification loss by up to 2% and achieving image inference speeds of just 1.4 milliseconds. This performance sets a new standard for efficient, real-time knife defect detection systems, promising cost-effective innovations in industrial automation testing.