Crack Detection Based on Multi-algorithm Fusion of Deep Learning and Image Processing

Aiming at the crack detection problem, this paper proposes an image processing technique based on Keras deep learning architecture and multi-algorithm fusion such as Retinex algorithm and wavelet transform to detect cracks. Considering the complexity of the actual image data, we first use principal component analysis, singular value decomposition and binarization operations to carry out preliminary processing of crack images. After obtaining the preprocessed data and inputting it into the model, the learning rate of 0.01, 0.001, 0.0001 and the number of iterations of 300, 500 and 700 are analyzed during the training process to obtain the optimal learning rate and number of iterations of the model. In order to further improve the stability of image feature recognition, we use Retinex algorithm and wavelet transform for light adjustment and denoising of the image, which provides a better support for crack detection in different application scenarios. Finally, we detect cracks in different material flooring, and the experimental results show that the image preprocessing improves the robustness and generalization ability of the model, and the deep learning architecture based on the optimal learning rate and number of iterations also has a better feature extraction effect. In summary, the use of multi-algorithm fusion of deep learning and image processing has a better utility in crack detection, and the relevant methods can be continuously explored in the future.