The article presents a neural network for convolution of aerial survey images to search and localize objects. When developing a convolutional neural network for convolution of aerial survey images, it is advisable to use the power of cloud technologies, by deploying the CNN on a cloud server. In this article, to construct a convolutional neural network with a full-scale network strategy, we used ResNet, of which architecture is bas. For traditional convolutional functions, neural networks in the process of convolution are characterized by a local receptive field, which can lead to the generation of local features. Encoding long-range contextual information is not performed properly, and the resulting local features can lead to significant potential disagreements between the features under study, which correspond to pixels with the same tags, resulting in inconsistencies within the class. pixels, eventually leading to low recognition efficiency. To solve this problem, the article improved the convolutional neural network for convolution of aerial survey images.