Intelligent Detection Method of Forgings Defects Detection Based on Improved EfficientNet and Memetic Algorithm

In the process of production, automobile steel forgings are prone to various cracks, which affect product quality. At present, defects mainly detected by fluorescent magnetic particle inspection and manual inspection. Aiming at problems low detection accuracy efficiency in this method, an improved convolutional neural network model is proposed. The images two typical were intelligently inspected. Firstly, a deep learning with EfficientNet as backbone Feature Pyramid Network (FPN) fusion layer constructed. Secondly, order improve convergence speed accuracy, calculation method intersection over union improved, using Attention Mechanism. Finally, Particle Swarm Optimization algorithm (PSO) adaptive parameters introduced optimize hyperparameters network, image acquisition platform built for verification. mean Average Precision (mAP) best EfficientNet-PSO on validation set 95.69%. F1 score 0.94 FLOPs 1.86B. Compared other five models, effectively improves defect flange plate cylinder head, can meet requirements.