Enhancing Tensor Based Imputation with CNNS for Astronomical Imagery Data

Satellite imagery frequently contains incomplete or corrupted regions caused by factors such as sensor malfunction, cloud obstruction, or data transmission errors, which adversely affect the accuracy of environmental monitoring and geospatial analysis. This proposed model presents a convolutional neural network (CNN)-based inpainting framework designed to reconstruct missing areas in green-area satellite images. The model adopts an encoderdecoder architecture composed of convolutional layers with batch normalization, enabling effective feature extraction and spatial context understanding for high-quality restoration. To enhance the perceptual fidelity of reconstructed images, we employ a hybrid loss function that combines Mean Squared Error (MSE) with a perceptual loss derived from intermediate feature maps of a pretrained VGG16 network. Performance is quantitatively assessed using Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index (SSIM), demonstrating the model's ability to generate visually coherent and semantically meaningful reconstructions. The proposed method introduces a lightweight yet perceptually robust approach to image inpainting, offering a computationally efficient alternative to heavier generative models, making it suitable for real-time or resource-constrained remote sensing applications.