ACCoNet

ACCoNet performs salient object detection (SOD) in optical remote sensing images (RSIs) to identify and localize prominent objects such as land use patterns, vegetation cover, and other geographical features.

Key Features:

• Encoder-Decoder Architecture: Employs an encoder-decoder framework that extracts and refines multi-level feature maps from RSIs.
• Adjacent Context Coordination Modules (ACCoMs): Uses ACCoMs to coordinate features across levels by activating salient regions in encoder outputs and transmitting this information to the decoder.
• Local and Adjacent Branches: Each ACCoM contains a local branch that adaptively highlights salient regions and two adjacent branches that integrate global contextual information from neighboring levels.
• Bifurcation-Aggregation Block (BAB): Implements a bifurcation-aggregation block in the decoder that integrates two bifurcations to capture extensive contextual information and aggregate features for saliency prediction.

Scientific Applications:

• Environmental monitoring: Enables detection of salient objects in RSIs for monitoring vegetation cover and land cover changes.
• Urban planning: Supports identification of urban structures and land use patterns from optical RSIs.
• Disaster management: Facilitates rapid localization of salient features relevant to disaster assessment in RSIs.

Methodology:

The encoder processes input RSIs to generate multi-level feature maps, ACCoMs refine these maps by highlighting salient regions and integrating contextual information across levels, and the decoder uses the bifurcation-aggregation block (BAB) to aggregate features and produce saliency maps.