A lightweight CNN-LSTM hybrid model for land cover classification in satellite imagery
Satellite image classification has a significant impact in the fields of agriculture, urban planning, and environmental monitoring. However, traditional Convolutional Neural Networks (CNNs) require high computational demand, a large number of parameters, and a long training time for classification t...
Saved in:
| Main Authors: | , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Taylor & Francis Group
2025-12-01
|
| Series: | Geocarto International |
| Subjects: | |
| Online Access: | https://www.tandfonline.com/doi/10.1080/10106049.2025.2521828 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Satellite image classification has a significant impact in the fields of agriculture, urban planning, and environmental monitoring. However, traditional Convolutional Neural Networks (CNNs) require high computational demand, a large number of parameters, and a long training time for classification tasks. To address this issue, a hybrid model combining a lightweight CNN with a Long Short-Term Memory (LSTM) network is proposed. Initially, input images are preprocessed using resizing, a sharpening filter, and contrast enhancement. Then, the CNN component extracts spatial features from the images using a lightweight structure consisting of two convolutional layers with 0.202 million parameters. Next, the LSTM component is used for temporal feature extraction from the learned features. Finally, softmax is used to classify the different types of satellite images. The proposed method achieved 98.8% accuracy on the RSI-CB256 dataset and 99.9% on the EuroSAT dataset, with a training time of 3 milliseconds per epoch. |
|---|---|
| ISSN: | 1010-6049 1752-0762 |