Lightweight camouflaged object detection model based on multilevel feature fusion

Qiaoyi Li,Zhengjie Wang,Xiaoning Zhang,Hongbao Du
DOI: https://doi.org/10.1007/s40747-024-01386-3
IF: 6.7
2024-03-09
Complex & Intelligent Systems
Abstract:Abstract The intrinsic similarity between camouflaged objects and background environment impedes the automatic detection/segmentation of camouflaged objects, and novel network architectures for deep learning are promising to overcome this challenge and improve detection accuracy. However, these existing network architectures for distinguishing between camouflaged objects and their backgrounds do not account for the constraint of detection speed, which results in high computational complexity and the inability to meet the requirements of rapid detection. Therefore, based on the human visual system, this study proposes a single-stage lightweight camouflage object detection network using multilevel feature fusion, integrating features of various feature layers and receptive field sizes. Using three benchmark datasets for normal camouflaged objects, the lightweight network (LINet) model demonstrated an accuracy superior to those of six existing mainstream camouflaged object detection methods. Its detection speed, 126.3 frames per second, is significantly higher than those of the existing mainstream methods, enabling rapid detection with a maximum increase of 187.62%. The accuracy of LINet is the minimum and maximum for Resnet101 and Resnet152, respectively. These findings pave the way for diverse applications of camouflaged target detection algorithms.
computer science, artificial intelligence
What problem does this paper attempt to address?
### Problems the Paper Attempts to Solve The paper aims to address two main issues in Camouflaged Object Detection (COD): 1. **Improving Detection Accuracy**: The inherent similarity between camouflaged objects and their background environment makes it very difficult to automatically detect or segment camouflaged objects. Although existing network architectures have made some progress in distinguishing camouflaged objects from the background, they are highly complex and fail to adequately consider the requirements for detection speed. 2. **Enhancing Detection Speed**: Existing methods usually have high computational complexity, making them unable to meet the needs for fast detection. This is a significant limitation in practical applications, especially in scenarios requiring real-time processing, such as military target detection and medical diagnosis. To overcome these issues, the authors propose a lightweight single-stage camouflaged object detection network (LINet) based on multi-level feature fusion. By simulating the human visual system, this network fuses features of different levels and receptive field sizes, thereby significantly improving detection speed while maintaining high detection accuracy. Specifically, LINet achieves a detection speed of 126.3 frames per second on three benchmark datasets, representing a maximum improvement of 187.62% over existing mainstream methods. Additionally, LINet also outperforms six existing mainstream camouflaged object detection methods in terms of detection accuracy. ### Summary The main contribution of the paper is the proposal of a lightweight and efficient camouflaged object detection model. By employing multi-level feature fusion and simulating the human visual system, the model significantly enhances detection speed while maintaining high detection accuracy. This achievement lays the foundation for the widespread use of camouflaged object detection algorithms in various application scenarios.