This study investigates the use of lightweight convolutional neural networks for automated detection of drone signals captured by a software-defined radio-based electronic warfare framework and shows that compact models can achieve high detection accuracy while maintaining low computational requirements, making them suitable for embedded radio-frequency monitoring applications.
Abstract
The increasing use of first-person-view drones in modern conflicts has created a demand for compact and reliable detection systems capable of operating in complex electromagnetic environments. These drones continuously transmit video signals through onboard video transmitters, generating radio-frequency emissions that can be exploited for early detection. This study investigates the use of lightweight convolutional neural networks for automated detection of drone signals captured by a software-defined radio-based electronic warfare framework. Samples are converted into rasterized time-domain images, providing a computationally efficient input representation suitable for embedded systems. Several custom model architectures were designed and benchmarked in terms of accuracy, model size, and inference performance using a dataset containing approximately 40,000 labeled images. In addition to offline testing, the models were integrated into a GNU Radio signal processing chain for real-time evaluation. The results show that compact models can achieve high detection accuracy while maintaining low computational requirements, making them suitable for embedded radio-frequency monitoring applications. Compared with existing spectrogram-based RF detection methods, the proposed approach eliminates frequency-domain preprocessing and achieves comparable accuracy with significantly reduced computational cost.
The advancement of unmanned aerial vehicles (UAVs) or drones has brought about emerging security concerns related to the protection of the border and national security, necessitating the development of effective and efficient drone detection systems. Although drones have immense benefits in various applications, their misuse creates severe threats like smuggling, surveillance, and even hostile acts across borders. Conventional border protection systems, whether radar-based or manual, are less effective in detecting small drones flying at low altitudes due to their low radar cross-sections and irregular flight paths. This paper presents a visual drone aided by Artificial Intelligence. A Convolutional Neural Network (CNN) based detection system. The proposed system is to learn by using deep learning algorithms. the separation of drone objects from others non-drone Visual image data based objects. The CNN model is developed to learn hierarchical spatial features from a series of the model consists of several convolution and pooling layers, followed by fully connected layers. The confusion matrix used for binary classification. The performance of the model is evaluated by traditional evaluation metrics like the accuracy, precision, recall and F1 score, confusion matrix analysis was conducted. and training-validation curves. The experimental outcome shows that the proposed system has an overall classification The test set has 94% accuracy. The proposed system is highly reliable in detecting non-drone objects with a recall of 0.99 and has a precision of 0.85 for drone object detection, thus reducing false alarms. The training and validation curves show that the system has a stable learning process with good generalization capabilities and less overfitting. The proposed system shows that deep learning-based visual detection systems can greatly improve border surveillance systems by offering accurate and scalable drone detection. The proposed system can be used for intelligent security systems and shows the efficiency of artificial intelligence in improving modern border surveillance infrastructure.
Ankita Malviya, Rashmi Singh, D. Tiwari et al.· 2026 International Conferenc...· 0 citations
This study presents an optimized, cost-effective radar sensor system for real-time, and contactless driver drowsiness detection. The system consist on a 24 GHz frequency-modulated continuous Wave radar module integrated behind the central rearview mirror for highly sensitive analysis of driver micromovements, including head tilts, and facial dynamics. The core innovation lies in the edge-deployed sensor-specific adaptation, where a custom CNN-LSTM pipeline processes complex radar returns to estimate eye opening and mouth opening metrics without visual data. This pipeline enhances the system's ability to isolate subtle motion characteristics—such as range and velocity—improving the sensing fidelity crucial for fatigue monitoring. Deployed on a resource-constrained Raspberry Pi platform, the model demonstrates robust integration and efficiency. Experimental validation on 20 subjects shows excellent performance, achieving up to 98.1% accuracy for eye blinking, and 97.9% for spontaneous yawning. This radar-based approach advances automotive sensing by providing a privacy-preserving, high-performance solution for safety-critical applications.
María-José López, César Palacios-Arias, B. Ordoñez et al.· IEEE Sensors Letters· 0 citations
Recent years have witnessed significant interest in drone positioning due to the rapid development of aerial surveillance, autonomous navigation, and intelligent monitoring applications. Although deep learning models achieve high drone detection accuracy, most existing systems lack an integrated vision-based framework for estimating the real-world three-dimensional position of drones. This paper suggests a drone positioning system that incorporates a deep learning detection system with stereo-vision-based localization in order to overcome this limitation. To train a YOLOv8 detection model, the UAVfly dataset of 1005 annotated drone images is used. Images are all resized to 640 640 pixels and also normalized to facilitate constant training. Once the drone is detected, the predicted bounding boxes are used to extract the corresponding regions of interest from the stereo image pairs. The SURF (Speeded-Up Robust Features) algorithm is utilized to extract the features and match them in order to detect significant points in the chosen region of the images. A brute-force matcher is then used to match these features of the left and right stereo images to get the correct corresponding points. On these matched points, depth is computed from disparity. Lastly, a 3D position of the drone is calculated by averaging the coordinates of several matched feature points, and this allows for more precise localization. The accuracy, precision, recall, F1-score, intersection over union, and mean average precision of the proposed method are 0.930, 0.942, 0.933, 0.938, 0.873, and 0.934, respectively. These findings confirm the effectiveness of the proposed method for accurate drone detection and reliable three-dimensional localization.
Tianchi Zhao· Journal of engineering and a...· 0 citations
Experimental results demonstrate that the proposed sequential pipeline preserves high specificity while reducing missed detections compared to radar-only processing and indicates that sequential processing can offer a viable alternative to parallel fusion for edge-based drone detection under SNR and computing constraints.
Faizal Mohd Amin Sharifuldin, N. E. Abdul Rashid, Mohd Adli Md Ali et al.· Sensing and Imaging· 0 citations
Vision-based monitoring systems are increasingly adopted in indoor environments to enhance fire safety by leveraging existing surveillance infrastructure. However, practical deployment remains challenging due to the computational demands of deep learning models and the need to process multiple camera streams in real time on resource-constrained devices. This paper presents a multi-camera edge AI system for real-time smoke and fire recognition using lightweight deep learning models. The proposed system operates on a Raspberry Pi 5 and integrates seamlessly with standard IP cameras, enabling deployment without additional sensing hardware. A MobileNetV2-based classification model is employed for efficient frame-level recognition, while a YOLO-based detection model is used for comparative analysis. A latency-aware processing pipeline is further designed to support multiple camera streams under limited computational resources. Experimental results show that the proposed system achieving an F1-score of 0.925 while reducing inference latency by over 6× compared to deeper models. The system maintains real-time performance of up to 30 FPS for a single camera and stable operation for up to four simultaneous camera streams. These results demonstrate that lightweight models provide an effective trade-off between accuracy and efficiency, making them suitable for practical edge AI deployment in fire monitoring systems.
Truc Thi Kim Nguyen, Thanh Huy Phan, Kim Hoàng Nguyễn et al.· E3S Web of Conferences· 0 citations
Today’s transportation systems suffer from a high number of accidents caused by drowsy driving, so strong automated detection systems are required for implementation in the Intelligent Transportation Systems (ITS). This paper introduces an analytical framework, divided into three stages to enhance the real-time detection of drowsiness at video level based on modules and provides the experimental results to analyze the proposed framework. To overcome the problems associated with computation-intensive deep learning-based methods, the proposed method involves employing Complete Local Binary Pattern (CLBP) texture descriptors, Dim: 512 with a greedy Pearson de-correlated feature selection that results in a trash-free subset of 58 texture bins (58 discriminative bins+auxiliary predictors), so that the features involved are largely reduced to 61. A comprehensive benchmark of 28 Classifier Presets including 9 Classifier families is performed on the NTHU Drowsy Driver Detection Dataset (357 video-derived observations) via five-fold cross validation. The best configuration, which is the Quadratic Support Vector Machine (SVM) trained on top 59 dimensions of normalised CLBP features, produces 92.42% accuracy and 88.1% dimensionality reduction with a weighted F1-score of 92.13%, which is 1.13 percentage points better than the weighted F1-score of the full-feature baseline. This solution offers an alternative to deep convolutional solutions for drowsiness detection in ITS that is both simple to interpret and able to use less computation.