With the development of science and technology, computer vision technology is increasingly used in new product market testing. Computer vision technology can use image analysis to evaluate and optimize the appearance, functionality, user feedback and other aspects of new products, thereby improving the market competitiveness and user satisfaction of new products. However, computer vision technology still has many problems, and its accuracy is limited by data quality. If the data quality is poor, the recognition rate will decrease. Factors affecting data quality include image clarity, resolution, noise, illumination, occlusion, deformation, etc. And compared to humans, who can easily identify complex image and video content, it can be very difficult for computers. Computer vision technology needs to deal with diversity and variability issues such as different perspectives, scales, postures, backgrounds, categories, etc. Computer vision technology requires a lot of algorithmic and mathematical knowledge, and the complexity of the algorithm and the requirements for computing resources can be very high. Computer vision technology needs to solve specific problems for different tasks, such as target detection, image segmentation, scene understanding, etc., and different models and methods need to be designed and optimized. Computer vision technology also involves people's personal privacy and data security issues, and corresponding measures need to be taken to protect privacy and security. Computer vision technology may be used for monitoring, tracking, identification and other purposes, which may infringe on people's rights and freedoms. Therefore, this article combines the AlexNet network with the Grad-CAM layer to derive the GCCV-CNN algorithm, which uses global context information. The GCCV-CNN algorithm introduces the global context convolution (GCC) module on the basis of CNN, which can effectively utilize the global context information of the image and improve the expressive ability and robustness of the features. Reduce parameters and calculation amount. The GCCV-CNN algorithm implements the GCC module by using depth-separable convolution (DSC) and point-wise convolution (PC), which can greatly reduce parameters and calculations, improve the efficiency of the model, and adapt to images of different scales.