Scholarlink

<p>Remote sensing scene classification stands at the core of modern geospatial intelligence and plays an important role in understanding and real time analysis of high resolution satellite images. With the introduction of multi-resolution, multi-spectral image data, satellite image classification has become more and more challenging. The integration of deep learning techniques in this sector has enabled a breakthrough in recent times. Although these methods showed quite good results, optimum performance is yet to be achieved. In addition, these methods require a large computational cost, leaving room for further research. In this paper, we propose a novel framework that increases the efficiency of satellite image classification by enhancing the performance and reducing the computational time. The proposed model integrates four different CNN models namely ResNet50, DenseNet121, MobileNetV2 and EfficientNetB0 to extract features from complex satellite images. A feature selection method is then applied to select the most important features. Afterwards, our proposed custom classification network is used to classify the selected features. All the experiments have been conducted based on two popular satellite datasets i.e., EuroSAT and NWPU-RESISC45. To evaluate the performance and generalizability of the proposed framework, we compared our results with a number of state-of-the-art approaches. The results demonstrate that the proposed framework is able to outperform the state-of-the-art approaches by achieving higher accuracy with much less computational time.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

MCFFCNet: A multi-CNN feature fusion framework with custom classification network for efficient satellite image classification

  • 期刊论文

  • 发表日期：

    2026年1月16日

查看全文
  • S. M. Nafis Ahmed,
  • Nishat Tabassum,
  • Hosney Jahan

摘要

Remote sensing scene classification stands at the core of modern geospatial intelligence and plays an important role in understanding and real time analysis of high resolution satellite images. With the introduction of multi-resolution, multi-spectral image data, satellite image classification has become more and more challenging. The integration of deep learning techniques in this sector has enabled a breakthrough in recent times. Although these methods showed quite good results, optimum performance is yet to be achieved. In addition, these methods require a large computational cost, leaving room for further research. In this paper, we propose a novel framework that increases the efficiency of satellite image classification by enhancing the performance and reducing the computational time. The proposed model integrates four different CNN models namely ResNet50, DenseNet121, MobileNetV2 and EfficientNetB0 to extract features from complex satellite images. A feature selection method is then applied to select the most important features. Afterwards, our proposed custom classification network is used to classify the selected features. All the experiments have been conducted based on two popular satellite datasets i.e., EuroSAT and NWPU-RESISC45. To evaluate the performance and generalizability of the proposed framework, we compared our results with a number of state-of-the-art approaches. The results demonstrate that the proposed framework is able to outperform the state-of-the-art approaches by achieving higher accuracy with much less computational time.