Capsule defect detection method based on Improved Faster RCNN

The capsule defect detection has always been a challenging issue. In previous research, it has been difficult to strike a balance between accuracy and detection speed for the capsule defect detection system to meet the requirements of industrial production. In light of this, I introduce an innovative approach to identify defects in capsules using an Enhanced Faster RCNN framework. This method is not only high in accuracy, but also meets the real-time requirements of industrial detection. This method effectively identifies common capsule anomalies like scratches, compression marks, and faulty imprints. Ensuring capsule quality is pivotal in production processes. To align with the demands of real-time industrial inspection, I incorporate the MobileNetV1 network, replacing the Vgg16 network in the traditional Faster RCNN for feature extraction. Network parameter learning is achieved using the Adam algorithm, a substitution for the conventional Momentum algorithm. Furthermore, recognizing the limitations of Faster R-CNN in precisely localizing smaller objects, I enhanced its capabilities by integrating the Feature Pyramid Network (FPN). This adaptation optimized the utilization of finely detailed shallow features, thereby improving the detection performance of smaller objects.