Local Ability Research Articles

A semantic segmentation network based on CNN and Swin transformer for landslide detection

Abstract To address the issues of poor object localization, difficulty in object recognition, and inadequate segmentation performance of CNN-based semantic segmentation methods in landslide detection, a semantic segmentation model based on the Swin transformer and ResUNet architectures for landslide detection in remote sensing images is proposed in this paper. This method combines the global feature capabilities of the Swin transformer with the local feature extraction abilities of CNN and integrates our proposed RCCT module for landslide segmentation in remote sensing images. We apply this model to the identification and extraction of landslide disasters in Menyuan County and the Sanjiangyuan region of Qinghai Province, China, achieving a promising F1 Score of 81.91. Furthermore, we conducted a comprehensive evaluation of our proposed model on the BiJie Landslide dataset in comparison to four recently developed methods for landslide recognition. The experimental findings indicate the superior performance and effectiveness of our model over the comparative approaches.

Adaptive PSO-SO algorithm with Sobol sequence for aerodynamic physical parameter identification of projectiles

In the domain of aerodynamic physical parameter identification, conventional optimization algorithms are often limited by falling into local optima. To overcome this limitation, a novel adaptive PSO-SO algorithm based on Sobol sequences (SAPSO-SO) algorithm is proposed in this study. The algorithm integrates particle swarm optimization algorithms and snake optimization algorithms, utilizing Sobol sequences for initialization, which enhances the global search and local development ability of the algorithm by adaptively adjusting the inertia weights and learning factors. In addition, this study introduced a local optimal discriminant mechanism and a local search function to further enhance the optimization performance of the algorithms. In this study, the small interval constant method was used to subdivide the trajectory, relying on the three-degree-of-freedom ballistic model to identify the starting ballistic parameters and aerodynamic physical parameters of each small interval. The performances of the snake optimization algorithm, particle swarm optimization algorithm, C-K method, and SAPSO-SO algorithm in the identification of ballistic physical parameters were compared using the full ballistic simulation data of a high-speed rotating projectile as measurement data. The results show that the SAPSO-SO algorithm demonstrates excellent accuracy and effectiveness, especially in noisy simulation data, where its recognition accuracy is improved by 7.79% over the C-K method, highlighting its superior anti-noise performance and global optimization capability. It is comprehensively analyzed that the SAPSO-SO algorithm has strong global optimization potential in theory and shows a high degree of accuracy and stability in practical applications, independent of the selection of initial parameters.

An interpretable hybrid framework combining convolution latent vectors with transformer based attention mechanism for rolling element fault detection and classification

Failure of industrial assets can cause financial, operational and safety hazards across different industries. Monitoring their condition is crucial for successful and smooth operations. The colossal volume of sensory data generated and acquired throughout industrial operations supports real-time condition monitoring of these assets. Leveraging digital technologies to analyze acquired data creates an ideal environment for applying advanced data-driven machine learning techniques, such as convolutional neural networks (CNNs) and vision transformer (ViT) to detect faults and classify, enabling accurate prediction and timely maintenance of industrial assets. In this paper, we present a novel hybrid framework based on the local feature extraction ability of CNN with comprehensive understanding of transformer within a global context. The proposed method leverages the complex weight-sharing properties of CNNs and ability of transformers to understand the larger context of spatial relationships in large-scale patterns, making it applicable to datasets of varying sizes. Preprocessing methods such as data augmentation are used to train the model on the Case Western Reserve University (CWRU) dataset in order to increase generalization through computational efficiency. An average fault classification accuracy of 99.62% is accomplished over all three fault classes with an average time-to-fault detection of 38.4 ms. MFPT fault dataset is used to further validate the method with an accuracy of 99.17% for outer race and 99.26% for inner race. Moreover, the proposed framework can be modified to accommodate alternative convolutional models.

PS-Net: A Learning Strategy for Accurately Exposing the Professional Photoshop Inpainting.

Restoring missing areas without leaving visible traces has become a trivial task with Photoshop inpainting tools. However, such tools have potentially illegal or unethical uses, such as removing specific objects in images to deceive the public. Despite the emergence of many forensics methods of image inpainting, their detection ability is still insufficient when attending to professional Photoshop inpainting. Motivated by this, we propose a novel method termed primary-secondary network (PS-Net) to localize the Photoshop inpainted regions in images. To the best of our knowledge, this is the first forensic method devoted specifically to Photoshop inpainting. The PS-Net is designed to deal with the problems of delicate and professional inpainted images. It consists of two subnetworks: the primary network (P-Net) and the secondary network (S-Net). The P-Net aims at mining the frequency clues of subtle inpainting features through the convolutional network and further identifying the tampered region. The S-Net enables the model to mitigate compression and noise attacks to some extent by increasing the co-occurring feature weights and providing features that are not captured by the P-Net. Furthermore, the dense connection, Ghost modules, and channel attention blocks (C-A blocks) are adopted to further strengthen the localization ability of PS-Net. Extensive experimental results illustrate that PS-Net can successfully distinguish forged regions in elaborate inpainted images, outperforming several state-of-the-art solutions. The proposed PS-Net is also robust against some postprocessing operations commonly used in Photoshop.

Toward identity preserving in face sketch-photo synthesis using a hybrid CNN-Mamba framework

The synthesis of facial sketch-photo has important applications in practical life, such as crime investigation. Many convolutional neural networks (CNNs) based methods have been proposed to address this issue. However, due to the substantial modal differences between sketch and photo, the CNN’s insensitivity to global information, and insufficient utilization of hierarchical features, synthesized photos struggle to balance both identity preservation and image quality. Recently, State Space Sequence Models (SSMs) have achieved exciting results in computer vision (CV) tasks. Inspired by SSMs, we design a hybrid CNN–SSM model called FaceMamba for the Face Sketch-Photo Synthesis (FSPS) task. It includes an original Face Vision Mamba Attention for modeling in latent space using SSM. Additionally, it incorporates a general auxiliary method called Attention Feature Injection that combines encoding features, decoding features, and external auxiliary features using attention mechanisms. FaceMamba combines Mamba’s modeling ability for long-range dependencies with CNN’s powerful local feature extraction ability, and utilizes hierarchical features at the appropriate position. Adequate experimental and evaluation results reveal that FaceMamba has strong competitiveness in FSPS task, achieving the best balance between identity preservation and image quality.

DeConformer-SENet: An efficient deformable conformer speech enhancement network

The Conformer model has demonstrated superior performance in speech enhancement by combining the long-range relationship modeling capability of self-attention with the local information processing ability of convolutional neural networks (CNNs). However, existing Conformer-based speech enhancement models struggle to balance performance and model complexity. In this work, we propose, DeConformer-SENet, an end-to-end time-domain deformable Conformer speech enhancement model, with modifications to both the self-attention and CNN components. Firstly, we introduce the time-frequency-channel self-attention (TFC-SA) module, which compresses information from each dimension of the input features into a one-dimensional vector. By calculating the energy distribution, this module models long-range relationships across three dimensions, reducing computational complexity while maintaining performance. Additionally, we replace standard convolutions with deformable convolutions, aiming to expand the receptive field of the CNN and accurately model local features. We validate our proposed DeConformer-SENet on the WSJ0-SI84 + DNS Challenge dataset. Experimental results demonstrate that DeConformer-SENet outperforms existing Conformer and Transformer models in terms of ESTOI and PESQ metrics, while also being more computationally efficient. Furthermore, ablation studies confirm that DeConformer-SENet improvements enhance the performance of conventional Conformer and reduce model complexity without compromising the overall effectiveness.

Ameliorated Fick’s law algorithm based multi-threshold medical image segmentation

Medical image segmentation is a critical and demanding step in medical image processing, which provides a solid foundation for subsequent medical image data extraction and analysis. Multi-threshold image segmentation, one of the most commonly used and specialized image segmentation techniques, limits its application to medical images because it requires demanding computational performance and is difficult to produce satisfactory segmentation results. To overcome the above problems, an ameliorated Fick's law algorithm (MsFLA) for multi-threshold image segmentation is developed in this paper. First, an optimized sine–cosine strategy is introduced to extend the molecular diffusion process to alleviate the problem of easily falling into local optima, thus improving the convergence accuracy of the Fick's law algorithm (FLA). Secondly, the introduction of local minimal value avoidance enriches the individual molecular information and enhances the local search ability, thus improving computational accuracy. In addition, the optimal neighborhood learning strategy is added to ensure a more careful and reasonable reliance on the optimal solution, thus reducing the chance of convergence of a local solution. The efficient optimization capability of MsFLA is comprehensively validated by comparing MsFLA with the original FLA and other algorithms in 23 classical benchmark functions. Finally, MsFLA is applied to image segmentation of grayscale images of COVID-19 and brain and color images of Lung and Colon cancer histopathology by using Cross entropy to validate its segmentation capability. The experimental results show that the MsFLA obtains the best segmentation results in three medical image cases compared to other comparison algorithms, which indicates that MsFLA can effectively solve the multi-threshold medical image segmentation problem.Graphical abstract

An autoencoder model based on one-dimensional neural network for epileptic EEG anomaly detection

We propose an autoencoder model based on a one-dimensional convolutional neural network (1DCNN) as the feature extraction network for efficient detection of epileptic EEG anomalies. The local information of normal EEG signals was captured by utilizing the local feature extraction ability of 1DCNN for training of an autoencoder to learn the expression of normal EEG data in low dimensional feature space. With the difference between the input and output as the anomaly score, the threshold was determined by the optimal equilibrium point of the ROC curve, and the EEG signals exceeding the threshold were diagnosed as the seizure data. The performance of the 1DCNN-AE epilepsy detection model was evaluated using the publicly available CHB-MIT scalp EEG dataset and TUH scalp EEG dataset. The AUC of the 1DCNN-AE model reached 0.890 of CHB-MIT and 0.686 of TUH under the average level of patients, and the epilepsy detection rate reached 0.974 and 0.893, and these results were better than the latest epilepsy anomaly detection models LSTM-VAE and GRU-VAE. The 1DCNN model had a parameter quantity of 58.5M, which was at the same level with LSTM-VAE (47.4 M) and GRU-VAE (36.9 M) but with much smaller FLOPs (0.377 G) than LSTM-VAE (21.6 G) and GRU-VAE (16.2 G). The autoencoder model based one-dimensional convolutional neural network can effectively detect abnormal EEG signals in epileptic seizure.

Mapless navigation via Hierarchical Reinforcement Learning with memory-decaying novelty

Hierarchical Reinforcement Learning (HRL) has shown superior performance for mapless navigation tasks. However, it remains limited in unstructured environments that might contain terrains like long corridors and dead corners, which can lead to local minima. This is because most HRL-based mapless navigation methods employ a simplified reward setting and exploration strategy. In this work, we propose a novel reward function for training the high-level (HL) policy, which contains two components: extrinsic reward and intrinsic reward. The extrinsic reward encourages the robot to move towards the target location, while the intrinsic reward is computed based on novelty, episode memory and memory decaying, making the agent capable of accomplishing spontaneous exploration. We also design a novel neural network structure that incorporates an LSTM network to augment the agent with memory and reasoning capabilities. We test our method in unknown environments and specific scenarios prone to the local minimum problem to evaluate the navigation performance and local minimum resolution ability. The results show that our method significantly increases the success rate when compared to advanced RL-based methods, achieving a maximum improvement of nearly 28%. Our method demonstrates effective improvement in addressing the local minimum issue, especially in cases where the baselines fail completely. Additionally, numerous ablation studies consistently confirm the effectiveness of our proposed reward function and neural network structure.

A Multi-Strategy Improved Northern Goshawk Optimization Algorithm for Optimizing Engineering Problems.

Northern Goshawk Optimization (NGO) is an efficient optimization algorithm, but it has the drawbacks of easily falling into local optima and slow convergence. Aiming at these drawbacks, an improved NGO algorithm named the Multi-Strategy Improved Northern Goshawk Optimization (MSINGO) algorithm was proposed by adding the cubic mapping strategy, a novel weighted stochastic difference mutation strategy, and weighted sine and cosine optimization strategy to the original NGO. To verify the performance of MSINGO, a set of comparative experiments were performed with five highly cited and six recently proposed metaheuristic algorithms on the CEC2017 test functions. Comparative experimental results show that in the vast majority of cases, MSINGO's exploitation ability, exploration ability, local optimal avoidance ability, and scalability are superior to those of competitive algorithms. Finally, six real world engineering problems demonstrated the merits and potential of MSINGO.

Fault Diagnosis of Distributed Energy Distribution Network Based on PSO-BP

With the increasing scale of distribution network at distribution time, its complexity grows geometrically, and its fault diagnosis becomes more and more difficult. Aiming at the slow convergence and low accuracy of traditional backpropagation neural network in dealing with single-phase ground faults, the study proposes a backpropagation neural network based on improved particle swarm optimization. The model optimizes the weights and acceleration constants of the particle swarm algorithm by introducing dynamic coefficients to enhance its global and local optimization seeking ability. It is also applied in optimizing the parameters of backpropagation neural network and constructing the routing model and ranging model for fault diagnosis about distributed energy distribution network. The simulation results revealed that the maximum absolute error of the improved method is 0.08. While the maximum absolute errors of the traditional backpropagation neural network and the particle swarm optimized backpropagation neural network were 0.65 and 0.10, respectively. The fluctuation of the relative errors of the research method was small under different ranges of measurements. At 8.0 km, the minimum relative error was 0.39% and the maximum relative error was 2.81%. The results show that the improved method proposed in the study significantly improves the accuracy and stability of fault diagnosis and localization in distribution networks and is applicable to complex distribution network environments. The method has high training efficiency and fault detection capability and provides an effective tool for distribution network fault management.

Improved Binaural Hearing Effects and Sound Source Localization in Listeners with Simulated Unilateral Conductive Hearing Loss: Effect of a Bone Conduction Device.

Objectives: It has been proven that patients with unilateral conductive hearing loss (UCHL) may encounter typical problems associated with asymmetric hearing, especially in challenging listening environments. In this study, we aimed to determine how UCHL affects speech recognition under multisource competing environments and the ability of sound source localization, as well as whether assistance with a bone conduction device (BCD) can confer hearing benefits in such listening tasks. Design: Acquired UCHL was simulated using an earplug combined with an earmuff in 10 listeners (mean age: 29.9 ± 4.77 years) with bilateral normal hearing (NH), and a within-subject repeated-measures design was used to compare hearing results among three listening conditions: NH (both ears open, C1), unilateral plug (UP) (simulated UCHL, C2), and UP + BCD (simulated UCHL aided with a BCD, C3). The speech reception threshold (SRT) of summation, squelch, and head shadow (HS) effects and the mean absolute error of sound source localization were used as markers for binaural hearing abilities. Results: BCD assistance (C3) improved the summation and HS effects in all listeners with simulated UCHL, resulting in a lower (i.e., better) SRT than that observed in C2; however, no significant differences in squelch effects were observed between C2 and C3. Notably, most listeners exhibited more accurate sound source localization in C3 than in C2. Further, BCD assistance mainly improved localization accuracy when the noise stimuli were presented at low intensities and on the hearing-impaired (plugged) side, suggesting that the benefits of BCD for sound localization are not based on the reacquisition of binaural processing. Conclusions: The current results have clinical implications for the promotion of BCDs in patients with UCHL, especially those with acquired UCHL who are unable to undergo surgery.

Energy-Saving Optimization of HVAC Systems Using an Ant Lion Optimizer with Enhancements

The complex and time-varying external climate conditions and multi-equipment variable coupling characteristics make it challenging to optimize the Heating, Ventilation, and Air Conditioning (HVAC) systems in existing buildings effectively. Additionally, the intricate energy exchange processes within HVAC systems present difficulties in developing accurate and generalizable energy consumption models. In response to these challenges, this paper proposes an Ant Lion Optimizer with Enhancements (ALOE) that can dynamically adjust the number of populations and the movement trend to improve the convergence speed and optimization ability, and randomly adjust the movement amplitude to enhance the local optimal escape ability. Finally, a case study of an office building in Hangzhou was carried out, and an overall energy consumption model of the HVAC system based on parameter identification and a general mechanism model was established. In this model, the energy-saving optimization effects of various advanced swarm intelligence optimization algorithms were compared. The experimental results demonstrate that under high, medium, and low load conditions, the ALOE algorithm achieves energy-saving rates of 28.16%, 28.26%, and 24.85%, respectively, the overall energy-saving rate for the entire day reaches 29.06%, which indicates the ALOE has significant superiority. This work will contribute to the development of energy-saving and emission-reduction technologies.

Local Management of Migrant Death in South Texas: A Necropolitical Landscape

We share the results of interviews and participant observations conducted in South Texas in spring of 2020 and summer of 2021 to understand how local authorities manage the issue of migrant deaths. Drawing on fifty-seven interviews and observations of local stakeholders from law enforcement and first responders to nonprofit organizations, United States Border Patrol (USBP) agents, and others, three experiences emerged. First, USBP both capacitates and limits local ability to respond to migrant deaths. Second, local jurisdictions feel a sense of impotence as a consequence of overwhelming migrant decedent caseloads, often leading to what we have termed an occupational paralysis. Third, failing to recognize migrant deaths as a regional phenomenon throughout South Texas hinders interior counties from receiving needed support. In our discussion we reflect on the experiences our interlocutors shared with us and the degree to which long-standing U.S. prevention through deterrence policy has configured the Texas–Mexico border into a necropolitical landscape, in the sense that U.S. Southwestern border enforcement, infrastructure, and the environment itself have a spatial arrangement that relegates migrant lives “outside the normal state of law” (Mbembe, 13), and funnels them toward death.

Experimental and Numerical Investigation of Acoustic Emission Source Localization Using an Enhanced Guided Wave Phased Array Method.

To detect damage in mechanical structures, acoustic emission (AE) inspection is considered as a powerful tool. Generally, the classical acoustic emission detection method uses a sparse sensor array to identify damage and its location. It often depends on a pre-defined wave velocity and it is difficult to yield a high localization accuracy for complicated structures using this method. In this paper, the passive guided wave phased array method, a dense sensor array method, is studied, aiming to obtain better AE localization accuracy in aluminum thin plates. Specifically, the proposed method uses a cross-shaped phased array enhanced with four additional far-end sensors for AE source localization. The proposed two-step method first calculates the real-time velocity and the polar angle of the AE source using the phased array algorithm, and then solves the location of the AE source with the additional far-end sensor. Both numerical and physical experiments on an aluminum flat panel are carried out to validate the proposed method. It is found that using the cross-shaped guided wave phased array method with enhanced far-end sensors can localize the coordinates of the AE source accurately without knowing the wave velocity in advance. The proposed method is also extended to a stiffened thin-walled structure with high localization accuracy, which validates its AE source localization ability for complicated structures. Finally, the influences of cross-shaped phased array element number and the time window length on the proposed method are discussed in detail.

Localization and Speech-in-Noise Performance in a Virtual Reality Test Environment: A Pilot Study of Adults With Single-Sided Deafness Using a Cochlear Implant.

The purpose of this exploratory study was to (a) construct a virtual reality (VR) test environment to measure speech recognition in noise (SIN) and localization, and (b) use the VR test environment to establish degree of binaural hearing benefit among a small number of adults with single-sided deafness (SSD) using a cochlear implant (CI). This pilot study included five adults implanted for SSD. The test environment was composed of an eight-speaker array that delivered restaurant noise and Institute of Electrical and Electronics Engineers sentences. VR head-mounted display goggles delivered a video recording of a busy restaurant. Participants completed SIN and localization in two conditions: (a) normal-hearing ear and a CI on the contralateral SSD side (CI-ON) and (b) normal-hearing ear and unaided on the contralateral SSD side (CI-OFF). Overall, CI benefits for SIN and localization within the VR test environment were improved for some participants, although not all. CI benefit for SIN and localization was dependent on speaker location. VR test environments present new opportunities for studying SIN and localization abilities in participants with CIs. This pilot study shows that, within a VR test environment, degree of CI benefit among SSD participants for SIN and localization varies across speaker location and across participants.

An improved Coati Optimization Algorithm with multiple strategies for engineering design optimization problems

Aiming at the problems of insufficient ability of artificial COA in the late optimization search period, loss of population diversity, easy to fall into local extreme value, resulting in slow convergence and lack of exploration ability; In this paper, an improved COA algorithm based on chaotic sequence, nonlinear inertia weight, adaptive T-distribution variation strategy and alert updating strategy is proposed to enhance the performance of COA (shorted as TNTWCOA). The algorithm introduces chaotic sequence mechanism to initialize the position. The position distribution of the initial solution is more uniform, the high quality initial solution is generated, the population richness is increased, and the problem of poor quality and uneven initial solution of the Coati Optimization Algorithm is solved. In exploration phase, the nonlinear inertial weight factor is introduced to coordinate the local optimization ability and global search ability of the algorithm. In the exploitation phase, adaptive T-distribution variation is introduced to increase the diversity of individual population under low fitness value and improve the ability of the algorithm to jump out of the local optimal value. At the same time, the alert update mechanism is proposed to improve the alert ability of COA algorithm, so that it can search within the optional range. When Coati is aware of the danger, Coati on the edge of the population will quickly move to the safe area to obtain a better position, while Coati in the middle of the population will randomly move to get closer to other Coatis. IEEE CEC2017 with 29 classic test functions were used to evaluate the convergence speed, convergence accuracy and other indicators of TNTWCOA algorithm. Meanwhile, TNTWCOA was used to verify 4 engineering design optimization problems, such as pressure vessel optimization design and welding beam design. The results of IEEE CEC2017 and engineering design Optimization problems are compared with Improved Coati Optimization Algorithm (ICOA), Coati Optimization Algorithm (COA), Golden Jackal Optimization Algorithm (GJO), Osprey Optimization Algorithm (OOA), Sand Cat Swarm Optimization Algorithm (SCSO), Subtraction-Average-Based Optimizer (SABO). The experimental results show that the improved TNTWCOA algorithm significantly improves the convergence speed and optimization accuracy, and has good robustness. Three‑bar truss design problem, The Gear Train Design Problem, Speed reducer design problem shows a strong solution advantage. The superior optimization ability and engineering practicability of TNTWCOA algorithm are verified.

Near Stochiometric LiNbO3 Crystal: the Piezoelectric Features and the Shear Horizontal Guided Wave Transducer for Structural Health Monitoring up to 650 °C.

The application of shear horizontal (SH) guided wave transducers in high-temperature structural health monitoring (SHM) is a topic of significant interest across various industrial engineering sectors. In this study, we utilized the novelty piezoelectric crystal of near stoichiometric lithium niobate (NSLN), which exhibited a robust piezoelectric response (d15 = 77.6 pC/N@room temperature). Next, the pure thickness shear vibration mode d15' through size optimization was designed. It was demonstrated that the NSLN-based ultrasonic guided wave transducers utilizing the optimum d15' mode were proficient in transmitting and receiving pure fundamental SH wave (SH0 wave) along two orthogonal main directions (0° and 90°) over a wide frequency range (100-350 kHz), exhibiting strong response to the SH0 wave. Under the driving voltage of 100 V, the signal voltages of the NSLN-based transducer were found to be on the order of 200.3 and 11.8 mV at room temperature and high temperature of 650 °C, respectively. Moreover, the NSLN-based SH0 transducer showcased its better defect localization ability, and the signal-to-noise ratio (SNR) sensitivity of NSLN-based transducer was evaluated to be 16.1 dB at high temperature of 650 °C. To sum up, the ultrasonic wave transducer based on NSLN crystal demonstrated higher potential applications for in situ SHM under elevated temperatures.

Sound Localization in Active Transcutaneous Bone Conduction Implant Users with Single-Sided Deafness.

Evaluate sound localization accuracy of subjects with single-sided deafness (SSD) with active transcutaneous bone conduction implants (atBCIs). Prospective case-control study. Tertiary referral center. Ten SSD patients (with ATBCIS) and 10 controls. Localization was assessed in a semianechoic chamber using a 24-speaker array. Stimuli included broadband noise (BBN) and narrowband noise (NBN). Perceived stimulus angle was recorded and compared with presented location. Statistical analyses were performed using ANOVA and Wilcoxon rank sum tests. The primary outcome measures were as follows: 1) mean angular error (MAE) error (°) and regression slope and 2) subjective benefit assessment (Speech Spatial Qualities questionnaire). Subjects with SSD demonstrated worse localization by MAE and regression slope compared with controls for both broadband noise (p < 0.0001) and narrowband noise at 500 Hz and 1000 kHz (p < 0.0001). There was no statistically significant difference (p = 0.1090) in slope between all groups at 4000 Hz. There was no significant difference in slope or MAE aided compared with unaided. Localization ability varied widely within the SSD cohort, with some individuals showing some ability in the unaided condition, best at 4000 Hz. Although SSQ confirmed particular difficulty in the spatial hearing domain, all domains improved with device use. Localization ability for individuals with SSD falls into a somewhat bimodal distribution. Some have fair localization, particularly at high frequencies, that is preserved but not improved with the atBCI. Others have minimal to no localization ability at any frequency, with no apparent device benefit.

An Improved Fire Detection Algorithm Based on YOLOv8 Integrated with DGIConv, FourBranchAttention and GSIoU

Fire detection is highly important for people's lives and property, and enhancing its accuracy is essential. This study focused on utilizing and improving YOLOv8 to obtain higher detection accuracy for fire detection. Three methods were used. First, the newly designed DGIConv module replaces the original Conv module, thereby decreasing the computational complexity while enhancing the model's performance. Second, to enhance the recognition ability of flame targets, a new attention mechanism named FourBranchAttention was designed, and a comparison was made with other attention mechanisms. The experiments revealed that the newly designed attention mechanism performed best on the mAP50 and mAP50-95 metrics. Finally, to improve the convergence speed and localization ability of the model, the loss function is optimized by adopting better hyperparameters of the TaskAlignedAssigner and employing the newly designed GSIoU as an alternative to the original CIoU. Through ablation experiments, all three improvements improved the detection performance to a certain extent, and the model using the three improvements achieved the best performance. Compared with the baseline, the YOLOv8 model with DGIConv, FourBranchAttention, and the optimized loss function increased the mAP50 by 2.52% and the mAP50-95 by 3.37%. The mAP50 and mAP50-95 had reached 98.46% and 75.26%, respectively. Compared with previous models, such as SSD/YOLOv7, the performance metrics of enhanced YOLOv8 also exhibited significant enhancements, thereby augmenting the accuracy of fire detection. Doi: 10.28991/HIJ-2024-05-03-09 Full Text: PDF