The development of tourism has been further enhanced by the growth of information technology and the provision of location-based services. One of the key infrastructures in location-based services for developing applications in recent years has been the recognition of tourist destinations and landmarks from images, enabling the delivery of relevant information to users. In this study, a processing web service is designed and tested to enable real-time recognition of landmarks from images. To achieve this, an image repository of famous landmarks in Tehran was prepared, and a conventional Convolutional Neural Network (CNN) was compared with a lightweight pre-trained CNN. The lightweight pre-trained CNN outperformed the conventional model, achieving an overall accuracy of 92% compared to 71%. Additionally, it showed superiority in other performance metrics and required significantly less training time—5 minutes versus 90 minutes. Following this, a web processing service was developed using TensorFlow and Flask and deployed on the Render cloud service provider. Time-based and scalability evaluations produced satisfactory results, showing a minimal latency increase of 0.039 in the presence of concurrent users. Furthermore, in 90% of tests, the server successfully responded. This research demonstrated that the proposed approach could serve as a suitable infrastructure for recognizing and retrieving information about tourist destinations in application development.