<p>As science and technology rapidly advance, the image recognition technology currently employed in the field of computer vision suffers from the drawback of inadequate recognition effectiveness. In response to this problem, an improved deep convolutional neural network model is introduced to recognize visual images to raise the precision of image recognition. The research improves the deep convolutional neural network model through wavelet analysis algorithm. The deep convolutional neural network model mainly reduces the number of parameters in image recognition, while the wavelet analysis algorithm mainly denoises the noise that appears in image recognition. On the basis of improving the deep convolutional neural network model, combined with the boundary Fisher analysis algorithm that can recognize high-dimensional data, an image recognition model is constructed to achieve efficient recognition of visual images. The experiment outcomes indicate that the proposed model has an average loss value of only 0.2573 when performing image recognition, significantly lower than other models, and its accuracy reaches 95.82%, significantly higher than other models. The proposed model achieves a recognition accuracy of 0.971 when recognizing images of different categories, significantly higher than other models. The above data indicate that the raised model has good recognition performance in the field of visual image recognition, and the image recognition accuracy is significantly higher than the other two models.</p>