MCFNet: Research on small target detection of RGB‐infrared under UAV perspective with multi‐scale complementary feature fusion
Authors:
Open Access
- 5 January 2025
- journal article
- research article
- Published by Institution of Engineering and Technology (IET) in IET Image Processing
- Vol. 19 (1)
Abstract
The task of single‐modal image target detection from a drone's perspective faces multiple challenges such as large image sizes, small and dense targets, insufficient lighting conditions, and hardware resource constraints, all of which affect the accuracy and real‐time performance of algorithms. In response, a small target detection algorithm that fuses multi‐scale complementary features, named MCFNet is proposed. Firstly, in order to independently extract the unimodal features of visible light and infrared images and effectively enhance the complementary information between images, a dual‐stream backbone network with a terminal fusion mechanism is proposed. Secondly, the complementary feature information residual module is utilized to optimize the integration process of residual features. Subsequently, by designing a multi‐scale feature enhancement module the network's capability to capture multi‐scale features is enhanced. Finally, for targets of varying sizes, a lightweight Transformer feature extraction module is proposed to improve the detection accuracy of small targets from a drone's perspective. Test results on the drone‐vehicle dataset show that this method achieved an average detection accuracy of 67.92%, while the detection accuracy on the self‐constructed UAV‐data dataset reached 96.4%. Additionally, a series of ablation experiments validated the effectiveness of the different modules.Keywords
This publication has 43 references indexed in Scilit:
- Modeling and compensation of small-sample thermal error in precision machine tool spindles using spatial–temporal feature interaction fusion networkAdvanced Engineering Informatics, 2024
- Predicting and compensating for small-sample thermal information data in precision machine tools: A spatial-temporal interactive integration network and digital twin system approachApplied Soft Computing, 2024
- Multi-source information fusion: Progress and futureChinese Journal of Aeronautics, 2023
- Combining transformers with CNN for multi-focus image fusionExpert Systems with Applications, 2023
- An improved fire detection approach based on YOLO-v8 for smart citiesNeural Computing and Applications, 2023
- A novel convolutional neural network based fault recognition method via image fusion of multi-vibration-signalsComputers in Industry, 2019
- FusionCNN: a remote sensing image fusion algorithm based on deep convolutional neural networksMultimedia Tools and Applications, 2018
- Multi sensor image fusion for surveillance applications using hybrid image fusion algorithmMultimedia Tools and Applications, 2017
- Multi-modal video event recognition based on association rules and decision fusionMultimedia Systems, 2017
- A review of remote sensing image fusion methodsInformation Fusion, 2016