Prefabricated Pizza Defect Detection Algorithm Based on CNN and Transformer

Accurate and efficient defect detection is of great significance in the field of prefabricated food quality inspection to ensure product quality and improve production efficiency. This article proposes a CNN Transformer model for defect detection in prefabricated pizza images. This method extracts three sets of pizza features by establishing a new color, texture, and shape feature extraction model, and uses three parallel convolutional neural networks for feature extraction. The extracted features are projected and input into a Transformer model combined with position encoding for processing. The projection section proposes a convolutional nonlinear projection method to improve the quality of projection. The position encoding part proposes a combination of sine cosine and learnable position encoding algorithms, utilizing the global position information provided by sine cosine position encoding and the adaptive adjustment ability of learnable position encoding to enhance the feature representation ability of the model. The Transformer encoder effectively captures long-range dependencies in the input sequence through multi-layer self-attention mechanisms and feedforward neural networks. Adaptive feature fusion integrates encoded features to achieve accurate classification of pizza defects.