The problem we aim to address is the challenge of real-time point cloud classification on resource-constrained devices, such as drones. To solve this, work proposes A Light weight framework for efficient point cloud classification using the dynamic graph convolutional neural network and mobileNetV3 (Point-DM). Point-DM integrates a Farthest Point Sampling module that samples the point cloud. This reduces the number of input points and significantly improves computational efficiency. The proposed Point-DM incorporates lightweight convolution blocks. These blocks progressively enhance feature representation. Additionally, MobileNetV3’s depthwise separable convolutions and attention mechanisms ensure both high classification accuracy and fast inference speed. Point-DM retains the edge feature extraction advantages of dynamic graph convolutional neural network. It also incorporates residual connections and a global aggregation module. These additions further strengthen feature expression capabilities while minimizing parameter usage. Comprehensive evaluation on the ModelNet40 dataset demonstrates that Point-DM achieves high accuracy. At the same time, it maintains outstanding efficiency. The results confirm the model’s strong scalability and broad application potential. Point-DM provides an efficient and reliable solution for 3D computer vision tasks on lightweight devices, such as drones.