An essential problem in computer vision and remote sensing, image-based scene classification has uses in anything from urban planning to environmental monitoring. Conventional Local Binary Patterns (LBP) are frequently employed for texture representation; however, their discriminative power in high-resolution satellite data is limited because they frequently miss color and spatial correlations. In order to effectively characterize local texture and scene semantics, this work suggests a Vector-Based Local Binary Pattern (VBLBP) descriptor that encodes both normalized spatial coordinates and binary color comparisons into a compact four-dimensional vector. The VBLBP characteristics are indexed using a k-d tree to enable neighborhood analysis and scalable retrieval. The VBLBP histograms are then used to train a 1D Convolutional Neural Network (1D-CNN) to learn high-level representations for scene classification. Tests on the PatternNet dataset, which comprises 30,400 high-resolution RGB images from 38 different scene classes, show that the suggested VBLBP-CNN framework performs better than other baseline techniques and conventional LBP variants, obtaining 94.6% classification accuracy while preserving computational efficiency. The complementary contributions of color and spatial components to the descriptor's performance are further supported by ablation investigations. The results demonstrate how deep learning models and manually created spatial-color descriptors can be used to provide reliable and scalable satellite scene classification.