Underground fires pose a significant threat to production safety in coal mines, and existing detection methods suffer from drawbacks such as poor adaptability to complex subterranean environments and excessive model parameters. To address the need for deploying object detection models on resource-constrained devices, this paper proposes a novel and efficient algorithm forUndergroundFireYOLOdetection, named UF-YOLO. The core innovation of this method is threefold: first, the StarNet module is introduced into the backbone to significantly reduce model parameters and computational complexity without sacrificing accuracy; second, the Cross-scale Context Fusion Module (CCFM) is integrated into the neck to enhance the model’s detection capability for fires of various scales, particularly small targets; and finally, Partial Convolution (PConv) is integrated to extract spatial features more efficiently, further reducing redundant computations and memory access. On our self-built Mine Fire Image Dataset (MFID), compared to the baseline model YOLOv11m, UF-YOLO reduces parameters by 77.1%, increases inference speed by 60.6%. Experimental results on the public COCO val 2017 dataset demonstrate that the proposed method outperforms state-of-the-art (SOTA) models such as YOLOv12. The results confirm that UF-YOLO can be efficiently deployed on the edge-side of coal mine IoVT monitoring systems to performe accurate and real-time fire detection. This work provides a new intelligent paradigm for the real-time monitoring of underground fires.

Recently, edge computing has emerged as a prevailing paradigm in applying deep learning-based object detection models, offering a promising solution for time-sensitive tasks. However, existing edge object detection faces several challenges: 1) These methods struggle to balance detection precision and model lightweightness. 2) Existing generalized edge-deployment designs offer limited adaptability for object detection. 3) Current works lack real-world evaluation and validation. To address these challenges, we propose theEdgeDetectionToolbox(ED-TOOLBOX), which leverages generalizable plug-and-play components to enable edge-site adaptation of object detection models. Specifically, we propose a lightweightReparameterized Dynamic Convolutional Network(Rep-DConvNet) that employs a weighted multi-shape convolutional branch structure to enhance detection performance. Furthermore, ED-TOOLBOX includes aSparse Cross-Attention(SC-A) network that adopts a localized-mapping-assisted self-attention mechanism to facilitate a well-craftedJoint Modulein adaptively transferring features for further performance improvement. Moreover, we propose anEfficient Headfor the classification and location modules to achieve more efficient prediction. Additionally, in practical industrial scenarios, we identify that helmet detection-one of the most representative edge object detection tasks-overlooks band fastening, which introduces potential safety hazards. To address this, we build aHelmet Band Detection Dataset(HBDD) and apply an edge object detection model optimized by the ED-TOOLBOX to tackle this real-world task. Extensive experiments validate the effectiveness of components in ED-TOOLBOX. In visual surveillance simulations, ED-TOOLBOX-assisted edge detection models outperform sixstate-of-the-artmethods, enabling real-time and accurate detection. These results demonstrate that our approach offers a superior solution for edge object detection.

Clear monitoring images are crucial for the safe operation of coal mine Internet of Video Things (IoVT) systems. However, low illumination and uneven brightness in underground environments significantly degrade image quality, posing challenges for enhancement methods that often rely on difficult-to-obtain paired reference images. Additionally, there is a tradeoff between enhancement performance and computational efficiency on edge devices within IoVT systems.To address these issues, we propose a multimodal image enhancement method tailored for coal mine IoVT, utilizing an ISP operations within a differentiable CNN framework fusion architecture optimized for uneven illumination. This two-stage strategy combines global enhancement with detail optimization, effectively improving image quality, especially in poorly lit areas. A contrastive language-image pretraining (CLIP)-based multimodal iterative optimization allows for unsupervised training of the enhancement algorithm. By integrating traditional image signal processing (ISP) with convolutional neural networks (CNN), our approach reduces computational complexity while maintaining high performance, making it suitable for real-time deployment on edge devices. Experimental results demonstrate that our method effectively mitigates uneven brightness and enhances key image quality metrics, with preservation of original visual information (PSNR) improvements of 2.9%–4.9%, structural similarity (SSIM) by 4.3%–11.4%, and visual information fidelity (VIF) by 4.9%–17.8% compared to seven state-of-the-art algorithms. Simulated coal mine monitoring scenarios validate our method’s ability to balance performance and computational demands, facilitating real-time enhancement and supporting safer mining operations.

For visual disbalance defects (VDDs) in low-light images, such as brightness unevenness and color imbalance, existing enhancement methods struggle to extract defect features from local regions and apply adaptive enhancement based on varying degrees of these defects. To address these challenges, we propose an unsupervised multi-modal enhancement method based on a high-order adaptive curve, named CLIP-AE. Specifically, we introduce a multi-modal recurrent optimization approach utilizing contrastive language-image pre-training (CLIP). This method iteratively optimizes variable embedded prompts and an Adaptive Enhancement Module (AEM) to establish dependencies between the prompts and detailed style features in the images, guiding the AEM to perform adaptive image enhancement. Additionally, we implement a progressive feature alignment strategy to enhance the model's ability to perceive style features and improve optimization efficiency by using multiple enhanced images with identical content features and incremental style features. In the AEM, the optimized Hyperparameters Generative Network (HGN) generates the optimal hyperparameters, which drive a High-Dimensional Nested Gamma correction (HDN-Gamma) to perform pixel-wise adaptive enhancement for VDDs. HDN-Gamma further maps pixel values using specific enhancement curves to avoid artifacts. Extensive experiments demonstrate that our method effectively improves visual disbalance defects and reduces artifacts. Compared to seven state-of-the-art algorithms, our method shows significant improvements (PSNR: 16.46%, 16.89%, and 15.14%; SSIM: 9.26%, 8.02%, and 9.85%; MUSIQ: 6.37%, 6.54%, and 7.45%) on the LOL, SICE, and MIT-Adobe FiveK datasets. Our approach offers a novel solution for applying multimedia technology in low-light image enhancement tasks.

Abstract Existing image enhancement algorithms often fail to effectively address issues of visual disbalance, such as brightness unevenness and color distortion, in low‐light images. To overcome these challenges, we propose a TransISP‐based image enhancement method specifically designed for low‐light images. To mitigate color distortion, we design dual encoders based on decoupled representation learning, which enable complete decoupling of the reflection and illumination components, thereby preventing mutual interference during the image enhancement process. To address brightness unevenness, we introduce CNNformer, a hybrid model combining CNN and Transformer. This model efficiently captures local details and long‐distance dependencies between pixels, contributing to the enhancement of brightness features across various local regions. Additionally, we integrate traditional image signal processing algorithms to achieve efficient color correction and denoising of the reflection component. Furthermore, we employ a generative adversarial network (GAN) as the overarching framework to facilitate unsupervised learning. The experimental results show that, compared with six SOTA image enhancement algorithms, our method obtains significant improvement in evaluation indexes (e.g., on LOL, PSNR: 15.59%, SSIM: 9.77%, VIF: 9.65%), and it can improve visual disbalance defects in low‐light images captured from real‐world coal mine underground scenarios.

With the rapid development of Internet of Things (IoT) and UAV technology, for the whole IoT system of vehicle detection, the middle and high level of information transmission and server processing has made a breakthrough, so at the bottom of the real-time detection of the vehicle by the UAV is the key to the whole system. However, UAV vehicle detection faces the challenges of too many small targets in the image leading to low detection accuracy, limited hardware platform resources requiring control of model size, and real-time detection requiring high inference speed. Aiming at the above problems, we propose a lightweight model PrFu-YOLO based on YOLOv8 improvement, which achieves a good balance between the accuracy, inference speed, and model size. And it realizes real-time vehicle detection embedded in an UAV platform. To solve the problem of low vehicle detection accuracy, we design a new structure PrFuFPN based on adding a small target detection layer to achieve more advanced feature fusion. To address the limited resources of the platform and the problem of real-time vehicle detection, we add GhostConv to the structure and constantly try to adjust the parameters of the network. Finally, extensive experiments were conducted on the VisDrone2019 and CARPK data sets to fully evaluate the model. Compared to YOLOv8s on the VisDrone2019 test set, mAP50 was improved by 10.05%, mA95 by 14.54%, the number of parameters was reduced by 8.13%, and the model size was reduced by 12.5%, while the FPS of 67.13 fully met the needs of real-time detection.

The power inspection is an important application of UAV-assisted power internet of video things (IoVT) for maintaining the safety of the power system. Due to the limitations of distance and angle, the resolution of the images captured by UAV is low, which seriously impacts the effects of small insulator defects detection. To address this problem, we propose a small-size defects detection method based on multi-scale feature interaction transformer for UAV-assisted Power IoVT. For the algorithm, we design a super-resolution reconstruction-assisted small object detection algorithm, the super-resolution module generates high-resolution images with the requirements of object detection function, which greatly improves the small object detection performance. Moreover, we design multi-scale feature interaction transformer network (MFITN), compared with the traditional non-local attention mechanism, the network structure can capture dependencies in multi-scales features, furthermore, the advantage assist the super-resolution module to generate more realistic image information to further improve small object detection. In addition, we propose a distributed model deployment strategy to deploy our high computational complexity algorithm in the edge side of the IoVT system, which can drive the overall algorithm to perform low-latency edge computation by relying only on the limited computing power devices. Experiments demonstrate that our method has better small object detection performance (mAP=81.3%, FPS=49.7), the super-resolution reconstruction is able to recover more realistic detail information, the distributed computing method can reduce the response latency by 33.4%-87.2%, which all contribute UAV-assisted Power IoVT system to realize accurate and fast power insulator defects detection.

Video surveillance as an important function of internet of video things (IoVT) system has been widely used in coal mine monitoring for coal mine safety with excellent results, however, there are still many shortcomings: 1) Existing coal mine IoVT systems have limited detection accuracy for small-sized objects; 2) Coal mine video surveillance systems generally adopt centralized cloud computing, transmission of massive data causes high latency, which seriously affects the response speed of object detection function; 3) The concept drift caused by the data stream seriously affect the detection effect of the offline algorithm. To address the above issues, we propose a small object detection method based federated intelligence to assist coal mine IoVT for object detection. First, we design a lightweight neural network Rep-ShuffleNet to improve YOLOv8, the state-of-the-art YOLO algorithm, to maintain high detection accuracy while dramatically increasing the inference speed, and with the advantage of lightweight, it can be deployed to embedded devices for low-latency edge computing; Moreover, we design a federated learning-based MLC-FL algorithm for local algorithms’ automatic and efficient optimization by asynchronous communication and data interaction reduction strategy. The experimental results show that with the assistance of federated intelligence model optimization strategies, the lightweight YOLOv8 has excellent detection performance (mAP: 94.6%, APsmall: 86.7%, FPS: 21.6), thus to assist coal mine IoVT to realize accurate and real-time underground small object detection.