In recent years, advancements in deep learning have significantly improved image retrieval systems, especially in mobile settings where computational resources are often limited. This review paper centers on deep learning methods designed specifically for image retrieval on mobile devices. The studies reviewed cover a range of techniques, including convolutional neural networks (CNNs), MobileNets, and contrastive learning, which aim to boost retrieval accuracy and efficiency. Key issues tackled include computational limitations, real-time processing, and the semantic comprehension of images. The research emphasizes the essential role of innovative optimization techniques and structural enhancements to fulfill the requirements of contemporary mobile applications. The findings highlight the necessity for lightweight designs and computational offloading strategies to effectively navigate resource constraints while upholding performance standards. Moreover, the paper delves into future opportunities in hybrid architectures, progressive learning frameworks, and methods for preserving privacy, outlining a path for continued advancements in mobile-focused image retrieval systems.