Precise land use classification from aerial data is essential for numerous applications, such as urban planning, monitoring of the environment, and management of resources. The field of artificial intelligence has evolved from conventional methods that use features to advanced deep learning techniques, notably Convolutional Neural Networks (CNNs). Despite the greater effectiveness of CNNs, issues remain in accurately capturing intricate spatial relationships, nuanced fluctuations in vision, and mitigating the lack of labeled data. This research presents a novel structure for land use classification that utilizes adaptive picture sharpening and a bespoke CNN architecture. Our methodology, Dynamic Gradient-Guided Sharpening (DGGS), selectively amplifies key characteristics in the images, so increasing their prominence for the model’s learning process. The enhanced imagery is utilized to train a bespoke CNN layer, combined with a pre-trained ResNet-152 architecture, to proficiently learn discriminative features for land use classification. We have experimented with the publicly available UC Merced Land Use Dataset and attain a benchmark accuracy of 99.33%, illustrating the effectiveness of our method in improving land use classification. This increases the efficacy of detection of land use in urban areas, reduces inaccuracies in environmental monitoring regarding deforestation, and optimizes the resource usage, such as precision agriculture. Our research findings contribute to better image interpretation by remote sensing and have enormous applications for numerous real-life applications. This directly enhances sustainable and informative decision-making processes.