A segmentation network for farmland ridge based on encoder-decoder architecture in combined with strip pooling module...

Area of Science:

• Agricultural remote sensing
• Computer vision for agriculture
• Precision farming technologies

Background:

• Wheat breeding and farmland management require accurate visualization of field characteristics like ridge size and spacing.
• Existing semantic segmentation methods struggle with ridge edge and strip feature detection in remote sensing imagery, limiting segmentation effectiveness.

Purpose of the Study:

• To develop an efficient and accurate method for farmland ridge segmentation using remote sensing data.
• To improve the extraction of ridge information for enhanced agricultural management and wheat breeding support.

Main Methods:

• Proposed a novel segmentation method employing an encoder-decoder network architecture.
• Integrated the Atrous Spatial Pyramid Pooling (ASPP) module for multi-scale context extraction.
• Incorporated a strip pooling module to enhance the capture of long-range dependencies in ridge features (both horizontal and vertical).
• Fused boundary and semantic features for final segmentation map generation.

Main Results:

• Achieved a high segmentation accuracy of 98.0% on the validation set.
• Obtained a mean Intersection over Union (mIoU) of 94.6%.
• Demonstrated precise segmentation of ridge information, outperforming conventional methods.

Conclusions:

• The proposed method effectively segments farmland ridge information, addressing limitations of existing techniques.
• The approach provides valuable data for farmland distribution analysis and supports practical applications in agriculture.
• This technology holds significant potential for advancing wheat breeding and precision farmland management.