Crop classification in high-resolution remote sensing images based on multi-scale feature fusion semantic...

Area of Science:

• Remote Sensing
• Computer Vision
• Agricultural Science

Background:

• Deep learning has shown great success in computer vision, offering opportunities for intelligent information extraction from remote sensing images.
• Deep convolutional neural networks are increasingly applied in agriculture for crop spatial distribution recognition.

Purpose of the Study:

• To address the challenge of improving crop classification accuracy and fine-grained image classification in remote sensing.
• To propose a novel multi-scale feature fusion semantic segmentation model (MSSNet) for enhanced crop recognition.

Main Methods:

• Crop mapping is framed as a semantic segmentation problem.
• The proposed MSSNet model utilizes multi-branch asymmetric convolution and dilated convolution for multi-scale feature extraction.
• Features are fused via concatenation, and skip connections integrate shallow and deep network features to enrich semantic information.

Main Results:

• Experiments using Sentinel-2 remote sensing images demonstrated that MSSNet effectively utilizes spectral and spatial crop characteristics.
• The model achieved good recognition effects, with improved plot segmentation and edge characterization of ground objects.
• The crop classification mapping output showed superior performance in detailed land cover delineation.

Conclusions:

• The MSSNet model offers a valuable reference for high-precision crop mapping and field plot extraction.
• This approach can help reduce excessive data acquisition and processing requirements in agricultural remote sensing.