Adjacent Context Coordination Network for Salient Object Detection in Optical Remote Sensing Images.

Area of Science:

• Computer Vision
• Remote Sensing Image Analysis
• Artificial Intelligence

Background:

• Salient object detection (SOD) in optical remote sensing images (RSIs) is crucial for image understanding.
• Existing SOD methods for natural scene images (NSIs) perform poorly on RSIs due to domain differences.

Purpose of the Study:

• To propose a novel network, ACCoNet, for improved salient object detection in optical remote sensing images.
• To address the limitations of directly applying NSI-SOD methods to RSIs.

Main Methods:

• Introduced the Adjacent Context Coordination Network (ACCoNet) with an encoder-decoder architecture.
• Developed Adjacent Context Coordination Modules (ACCoMs) to coordinate multilevel features and enhance salient regions.
• Proposed a Bifurcation-Aggregation Block (BAB) to capture contextual information within the decoder.

Main Results:

• ACCoNet demonstrated superior performance compared to 22 state-of-the-art methods across nine evaluation metrics.
• Achieved a high inference speed of up to 81 frames per second on a single NVIDIA Titan X GPU.
• Validated effectiveness on two benchmark datasets for RSI-SOD.

Conclusions:

• ACCoNet offers a significant improvement for salient object detection in optical remote sensing images.
• The proposed ACCoMs and BAB effectively leverage adjacent and multilevel features for enhanced detection.
• The method provides a computationally efficient and accurate solution for RSI-SOD.