This paper introduces a deep learning-based approach for point cloud classification, leveraging the PointNet architecture to optimize 3D object recognition. The method effectively addresses the challenges associated with unordered point cloud data, achieving superior classification performance with 92% accuracy, 91% precision and recall, 89% F1-score, and 96% sensitivity and specificity. The proposed model captures spatial features directly from raw point cloud data, demonstrating its potential for real-world applications in 3D object recognition and scene understanding. Comprehensive experiments on benchmark datasets validate the model’s effectiveness in classifying complex 3D structures, highlighting its robustness and efficiency. Future research will focus on advancing feature extraction techniques and refining the model to enhance classification performance under more demanding conditions.