acrobyte007/Optimiezed-CNN

Two lightweight CNN models, CS-CNN and SE-CS-CNN, were developed for efficient classification on low-resource edge devices. These models achieve approximately 94% accuracy on satellite imagery with a reduced parameter count of 1.25M for enhanced real-time performance. #**Satellite Image Classification Using Lightweight CNN Architectures** This project presents optimized Convolutional Neural Network (CNN) architectures designed for efficient and accurate classification of satellite images, particularly suited for deployment in real-time and low-resource environments like drones or edge computing platforms. **🚀 Features** Two lightweight CNN models: Channel Separated CNN (CS-CNN) Channel Separated CNN with Squeeze-and-Excitation (SE-CS-CNN) High classification accuracy (~94%) with minimal computational cost. Efficient use of parameters (~1.25M vs. 1.87M in standard CNNs). Adaptable to multispectral and derived indices (e.g., NDVI, NDBI). Compatible with small input image sizes (64x64x3). **🛠 Technologies Used** Python 3.12 TensorFlow / Keras NumPy, Pandas Scikit-learn (for preprocessing and evaluation)