Furniture Image Style Recognition Based on Neural Network

In order to solve the problem of fine-grained recognition of furniture style, the author proposes an improved bilinear CNN (rotational neural network) model. Introduce spatial attention mechanism into the output feature map of the VGGl6 feature extraction network to enhance attention to furniture areas in the image. Design a bilinear pooling method with a single feature path, which reduces the number of parameters and computation while fully mining fine-grained features using two operations: global average pooling and global maximum pooling. The experimental results show that the improved bilinear CNN model achieves a recognition accuracy of 76.4% on the FashionStylel4 dataset. Compared to the original bilinear CNN model, the improved model improves recognition accuracy by 2 percentage points, reduces parameter count by 40% and computational complexity by 52%, and has robustness in style recognition of furniture images with disturbances.