An Improved Encoder-Decoder Network Based on Strip Pool Method Applied to Segmentation of Farmland Vacancy Field.

Area of Science:

• Remote Sensing
• Agricultural Science
• Computer Vision

Background:

• Traditional remote sensing for crop analysis uses high-altitude imagery, which lacks detailed farmland insights.
• Semantic segmentation of high-altitude images provides regional crop coverage but not specific farmland conditions.

Purpose of the Study:

• To develop a more accurate method for segmenting farmland vacancies using low-altitude remote sensing imagery.
• To improve the extraction of vacancy features in agricultural fields through novel deep learning modules.

Main Methods:

• A new dataset was constructed using low-altitude farmland images.
• Several semantic segmentation algorithms were compared.
• Strip Pooling Module (SPM) and Mixed Pooling Module (MPM) were designed and integrated into a semantic segmentation network.
• An improved ResNet network and data augmentation were employed to enhance model performance and reduce annotation costs.

Main Results:

• The proposed method achieved 95.6% accuracy and 77.6% mean Intersection over Union (mIoU) on the test set.
• The model demonstrated a 7% error rate.
• The mIoU value showed an improvement of nearly 4% compared to existing models.

Conclusions:

• The developed method is highly effective for farmland vacancy segmentation using low-altitude imagery.
• The integration of SPM and MPM significantly enhances feature extraction for vacancy detection.
• The model's performance indicates its suitability for practical agricultural applications.