Search

Deep neural networks (DNNs) have driven the recent advances in fire detection. However, existing methods require large-scale labeled samples to train data-hungry networks, which are difficult to collect and even more laborious to label. This paper applies unsupervised domain adaptation (UDA) to transfer knowledge from a labeled public fire dataset to another unlabeled one in practical application scenarios for the first time. Then, a transfer learning benchmark dataset called Fire-DA is built from public datasets for fire recognition. Next, the Deep Subdomain Adaptation Network (DSAN) and the Dynamic Adversarial Adaptation Network (DAAN) are experimented on Fire-DA to provide a benchmark result for future transfer learning research in fire recognition. Finally, two transfer tasks are built from Fire-DA to two public forest fire datasets, the aerial forest fire dataset FLAME and the large-scale fire dataset FD-dataset containing forest fire scenarios. Compared with traditional handcrafted feature-based methods and supervised CNNs, DSAN reaches 82.5% performance of the optimal supervised CNN on the testing set of FLAME. In addition, DSAN achieves 95.8% and 83.5% recognition accuracy on the testing set and challenging testing set of FD-dataset, which outperform the optimal supervised CNN by 0.5% and 2.6%, respectively. The experimental results demonstrate that DSAN achieves an impressive performance on FLAME and a new state of the art on FD-dataset without accessing their labels during training, a fundamental step toward unsupervised forest fire recognition for industrial applications.

The training of video fire detection models based on deep learning relies on a large number of positive and negative samples, namely, fire video and scenario video with other disturbances similar to fire. Due to the prohibition of ignition in lots of indoor occasions, the fire video samples in the scene are insufficient. In this paper, a method based on generative adversarial network is proposed to generate flame images which are then migrated into specified scenes, thus increasing fire video samples in those restricted situations. Flame kernel is pre-implanted into the specified scene to keep its characteristics intact. The flame and scene are blended together by adding styling information such as blurry edge and ground reflection. This method overcomes background distortion which is caused by existing multimodal image translation on as a result of information loss and is able to guarantee the diversity of flames in specified scenes and produce perceptually realistic results. Compared with other multimodal image-to-image translation schemes, the FID and LPIPS values of images generated by our method are the highest, reaches 118.4 and 0.1322 respectively. In addition, Unet and the SA-Unet, in which a self-attention mechanism is involved, are used as fire segmenting networks to evaluate the enhancement of the augmented data on improving the accuracy of segmented network. Their F1-scores reaches 0.8905 and 0.9082 respectively after Unet and SA-Unet are trained with GAN-based augmented dataset generated by our model. The F1-scores are second only to 0.9259 and 0.9291 which are obtained when Unet and SA-Unet are trained with real picture serving as augmented dataset.

At present, all CNN-based fire detection algorithms identify fire by means of a single frame image, all of which demonstrate low accuracy under strong interferences or complex backgrounds such as flickering light or backgrounds with high level of brightness. To increase the accuracy of fire detection, this paper presents a neural network model which combines lightweight CNN with SRU. In this algorithm, the scene content is extracted by CNN and the dynamic characteristics of the flames are extracted from sequential frames. In this paper, Resnet18+SRU (V1-type) and Mobilenets+SRU (V2-type) are proposed. Based on the characteristics of flames at a fixed position within a short period of time, a 3D convolutional layer is added between the Mobilenets and the SRU in the V2-type model, resulting in the V3-type model. Based on a cross validation set containing multiple types of interference in an indoor environment, experiments were conducted to compare the three models proposed in this paper with other models. The experiment results showed that the accuracy of the method proposed in this paper is above 96%, about 25% higher than the accuracy of CNN-based fire alarm via single-frame image, and that the V3-type models with 3D convolutional layer has the highest accuracy and best overall performance.

The usually, softening occurs when metals are annealed, while for some special metals or alloys, the abnormal phenomenon of annealing hardening will occur. By means of X-ray diffraction (XRD), metallographic (OM) observation, transmission electron microscopy (TEM) and hardness test, the microstructure and hardness of 79% cold rolled Al0.5CrFeCoNiCu high entropy alloy was studied after isothermal annealing for 1 hour at different temperatures, which revealed the cause of annealing induced hardening of the alloy. The results show that the homogenized high-entropy alloy is a face centered cubic (FCC) biphasic structure with close lattice constant. After annealing at 300-750°C, the microhardness of the alloy is higher than that of the cold rolled alloy (399hv); With the increase of annealing temperature, the hardness increased first and then decreased. After annealing at 500 °C for 1 hour, the hardness reached the peak (489hv). XRD phase analysis shows that the alloy will precipitate body centered cubic (BCC) phase when annealed at 500-950°C and precipitate σ phase when annealed at 700-850 °C; In addition, when annealed at 650°C and above, the dislocation recovery speed in FCC phase is accelerated and recrystallization will occur, resulting in rapid softening of the alloy. TEM observation showed that there were L12 nano-precipitates with ordered structure in the FCC phase of the matrix. Therefore, the reason for the hardness peak of the alloy annealed at 500°C is the joint action of dislocation strengthening and precipitation strengthening (L12 and BCC phases).

In this paper, a video fire detection method is proposed, which demonstrated good performance in indoor environment. Three main novel ideas have been introduced. Firstly, a flame color model in RGB and HIS color space is used to extract pre-detected regions instead of traditional motion differential method, as it is more suitable for fire detection in indoor environment. Secondly, according to the flicker characteristic of the flame, similarity and two main values of centroid motion are proposed. At the same time, a simple but effective method for tracking the same regions in consecutive frames is established. Thirdly, a multi-expert system consisting of color component dispersion, similarity and centroid motion is established to identify flames. The proposed method has been tested on a very large dataset of fire videos acquired both in real indoor environment tests and from the Internet. The experimental results show that the proposed approach achieved a balance between the false positive rate and the false negative rate, and demonstrated a better performance in terms of overall accuracy and F standard with respect to other similar fire detection methods in indoor environment.

Fire is one of the most important factors that threaten the safe sailing of ships. Timely and accurate video fire detection relies on fast and accurate extraction of suspected flame regions. In view of the complicated environment of varied illumination and numerous interferences in the large ship space, this paper analyzed the different methods of object extraction based on four typical flame and interference video images. Experiments showed that the commonly used motion detection method is not suitable for the extraction of fire targets in a large ship space, while the RGB-HIS flame color model proposed by Chen et al. has a good performance and can be used in the extraction of suspected flame regions.

In order to solve the problem of insufficient flame image data, this paper designs a flame image data sets generation algorithm based on generative adversarial network and studies the influence of different generator depth and training times on the flame generation effect. The effect of generative flame under different conditions was quantitatively evaluated by two indexes of Inception score and PSNR. The simulation results show that when the generator is 9 layers and training is 80 times, the Inception score is 2.06 and PSNR is 13.18. The effect of generative flame image is better at this time. Therefore, this generative adversarial network model can better realize the purpose of generating the flame image data sets.

The rapid spread of the coronavirus (COVID-19) has led the world into a huge pandemic crisis. As of November, more than 50 million people have been infected. In response to the transmission characteristics of COVID-19, the World Health Organization (WHO) has identified wearing masks as the most effective anti-epidemic measure in public places and crowded places. In order to ensure that people entering these areas wear masks, this paper designs a complete epidemic prevention system. Utilizes deep learning, embedded technology and LoRa wireless communication. It is characterized by low cost and flexible communication methods, which can well meet actual epidemic prevention needs.

With the continuous development of China’s economy, people’s quality of life is constantly improving. As a matter of course, more and more attention is paid to home design. In particular, the space design of small apartments is welcomed by everyone. Among them, soft cloth decoration is a very popular form of home design now, which is very suitable for the design and use of small apartment. But the diversity of small family space brings difficulties to the design. The design and calculation of the simple home space in the small apartment with the RANSAC algorithm is fully satisfied with the requirements of the use of the algorithm and can achieve the accuracy of our calculation and facilitate communication with customers. The RANSAC algorithm has the powerful ability of image processing, which can help us to establish the visualization research of space design.

At present, all CNN-based fire identifications identify whether a fire is blazed with a single frame image, all of which have low accuracy under a strong interference or complex backgrounds such as flickering light or a high-brightness background. To address the issue, this paper proposes a neural network model which combines CNN with SRU. The scene content is extracted through CNN and combined with the dynamic characteristics of the flame extracted by SRU to improve the accuracy of fire alarm. There are three models proposed in the paper, including Resnet18+SRU, Resnet34+SRU and resnet18(Maxpool)+SRU. The models were validated on a test set containing intensive indoor environmental interferences and compared with CNN-based single-frame and multi-frame fire identification methods. The results show that the proposed methods are more accurate than the single-frame CNN fire recognition method by more than 25% and are more suitable for indoor environment fire alarms.

In the field of video-based fire detection, traditional fire color models have poor adaptability and susceptibleness to environmental interference. As fire can be divided into fire core, inner fire and outer fire layers depending on its temperature and color, a new fire detection model based on dispersion of fire color components has been proposed by the author. In order to further reduce false alarm, similarity of consecutive video frames of detected area is incorporated into the method, and a tracking algorithm based on flame kinematic characteristics of the detected area is proposed. Experimental results demonstrated that the proposed fire detection method can eliminate the influence of regular interference sources and accurately generate fire alarms, and exhibits practical values for indoor fire detection.

In the field of video fire detection, traditional fire color models have the poor adaptability and week robustness to interference. As fire can be divided into fire core, inner fire and outer fire areas depending on it's temperature and color, a new fire recognition model referring to the dispersion of fire color components is proposed in this paper. The threshold of B component standard deviation is calculated out by drawing the ROC curve of detecting results based on large number of sample images. A series of experiment results show that the proposed color model can eliminate the influence of interferences and noises, and detect out the fire areas accurately in the image, which has a certain practical significance for fire detection.

This paper presents an effective method based on support vector machines for identification of power system disturbances. Because of its advantages in nonlinear systems applications, phase space reconstruction is used to extract the distinctive features of the voltage signals. Eighteen particular disturbance signals are selected and then analyzed using this tools. Phase space reconstruction and support vector machines are used to identify and classify a variety of power disturbance signals. The results show that the methods proposed in this work are effective.

DC micro-grid has a number of distributed energy resources and a large number of constant power loads. They are prone to oscillate, bringing hidden hazards to the stable operation of DC micro-grid. With the example of two-source and two-load cascade system coupling, this paper discusses about one strategy to inhibit DC micro-grid oscillation. The strategy is to introduce the heterogeneity coefficient of line-regulating converter, combined with values of coupling control parameter; The feasibility of both methods is verified by simulation analysis.

Obstacle detection is a key technology of intelligent transportation and autonomous robot navigation. Aiming at the shortages of traditional obstacle detection technologies, the paper applies the Kinect depth camera as the sensor of obstacle detection system, and an obstacle detection method based on Kinect depth image is proposed on the theoretical basis of Kinect real-time 3D reconstruction and Meanshift segmentation. A series of experiments are carried out in different scenes, and the results show that the method can detect obstacles exactly in complex environments and it's robust to the affects of illumination, so it has a certain practical significance for the research of obstacle detection technology.

With the wide application of power electronic equipment, more and more attention has been paid to the power quality of power supply system. For three-phase/switch VIENNA rectifier with a high power factor, one cycle control is a reliable and simple non-linear large signal control method. For the design of VIENNA rectifier, the balance of dc-link neutral-point voltage is important, especially when DC side is connected to unbalance load. This paper puts forward a new neutral balance control algorithm based on one-cycle control of VIENNA rectifier and improves the current three-phase/switch one-cycle control algorithm, making it fit with the realization of discretization.

The hardware implementation scheme and software processing flow of IEEE1588 clock synchronization techniques in the Ship Power Synchronized Phasor Measurement(SPSPM) was discussed in this paper. The technology of Frequency compensated clock was adopted as Frequency self-compensation scheme for the problem that Clock drift deviation expand gradually in the synchronization interval. The scheme was test on the PMU hardware platform in SPMU, the test results show that the steady-state accuracy of synchronization are within ±100ns.

In the variable frequency driver system of induction motor, the DC-side current of converters and the AC-side current of rectifiers are analytically analyzed using switching functions of converters and rectifiers when induction motors would be in fault state. The transmission rules of faults characteristic information in driver system were derived, and the fault features frequency components corresponding to induction motor faults are acquired in the spectral of the DC-side current of converters and the AC-side current of rectifiers. In terms of the conclusion as mentioned, take the fault diagnosis of the broken rotor bars faults as an example to explain the transmission rules of faults characteristic information based on the driver system. Experimental results show that the theoretic analysis are correct, and the research provides theoretic foundation for diagnosing motor faults based the rectifier's AC- side current spectral analysis technology.

A method based on the D-S (Dempster-Shafer) evidential theory is presented and applied to diagnose the broken rotor bars of induction motors. Wavelet packets decomposition of the three-phase stator current is performed based on the frequency partition characteristics of the wavelet packets transformation; The faults characteristic vector (evidence) of the broken rotor bars is established by the RMS (Root Mean Square value) value of the wavelet nodes coefficient; Minkowaki distance measurement is used to construct basic reliability distribution function and to obtain its value which is is evidence as a result of the rotor faults. With the fused processing according to the D-S evidential fusion algorithm, the broken rotor bar faults of the motor is accurately identified. The experimental results demonstrated that reliable diagnosis of the broken rotor bars can be achieved by the proposed method.

Firstly this paper presents the technical aspects of designing a Dynamic Voltage Restorer (DVR) to meet some stringent requirements of voltage sags mitigation for Micro Integrated Power System (MIPS). Secondly through an analysis of configuration of DVR the optimum energy injection scheme is introduced. Thirdly the principles of a control scheme and a protection circuit for mitigating surge current in case of switching on DVRs are demonstrated. The experimental results from a prototype under some conditions of voltage sags are demonstrated to verify a good system performance.

A new method of data interaction between C# program and Simulink model is presented for the power systematic three-dimensional simulated training system. By establishing shared memory to be medi-location of data, Simulink model read or write from share memory through program in S-function. C# program uses Win32 API function to accomplish writing programmed segment for establishing shared memory and controlling Simulink model starting. Thus, the interaction of data is realized. Experiments demonstrated this method offered a new mix program solution to take each advantage of C# Language and Matlab.

According to the feature of DC superposition theory which is applied widely in monitoring the insulation condition of power cable with cross-linked polyethylene insulation (XLPE), such as high precision but time-consuming, so a quick positive and negative DC superposition monitoring method is presented in this paper. In which the Positive and negative DC voltages are injected to detection circuit respectively, and the steady voltages of the two conditions mentioned above are calculated from the transient voltages of sampling resistance. Considering the algebraic average of positive and negative steady voltages as finally steady voltage, so the value of insulation resistance for XLPE power cable is obtained. The proposed method not only has avoided infection in measuring precision resulted from water treeing, but also taken into account distributing capacitance; the monitoring time has been decreased. WINBOND W78E58B has been used in signal processing and the experiment results have affirmed the availability.

In order to reduce false alarm and alarm failure of fire detection system, a research of fire detection method based on multi-sensor data fusion was conducted. In this research, probabilistic neural networks (PNN) data fusion algorithm was employed to detect fire based on texture features from fire scene. Information of temperature and smoke concentration were processed by trend algorithm separately. Results from the above three fire detection algorithms were processed through decision level data fusion to accomplish fire detection and automatic fire alarm. It has been demonstrated that fire detection platform based on this method can detect fire faster and more accurately and discard nuance disturbances from florescent light or alcohol burner, thus providing a brighter future in applications.

Package is a critical part in biomedical implantable systems. Many factors affecting the host body and the life time of implantable systems need to be considered. Package becomes more critical for microfabricated systems with wireless charging and communication. This paper presents the first phase study on micro package techniques for short term (30 to 90 days) implantable systems. A MEMS implantable telemetry model system was designed for packaging evaluation. The transmitter was custom designed and fabricated using MOSIS processes and an external receiver was designed and built for data collection. For short term implantable systems, medical grade silicone outer coating is used for "tissue compatibility"; while multilayer polymeric and nanometer-thin metal or ceramic films were used for inner coatings to provide mechanical strength and to block vapor and moisture penetration. The total coating thickness is less than 0.6 mm. The electrical performances (leakage resistance) of test board and model devices coated with various package materials and processes are evaluated in 40 degrees C saline. This paper presents: the model system; the evaluation methods and analysis of failure modes of polymeric coating on test boards; the solution to the failures and suggested coating techniques of polymeric materials; and the evaluation of model systems packaged with multi-layer coatings in 40 degrees C saline. The expected performance of developed packaging method was verified by experiments. Implantable wireless MEMS system can be packaged with thin multilayer materials to have an expected life time greater than 30 days.

A crack propagation behavior of nano size hcp crystal (13nm × 24nm × 4.6nm) has been simulated by molecular dynamics method using Lennard-Jones type potential. In a case of model crystal with initial notch plane and direction were (101 0), [12 10], the crack propagated parallel to notch plane and two sets of prismatic slips were occurred at the crack. Therefore, the crack in this crystal is deduced to extend by alternating shear on two intersecting {101 0} prismatic slip systems. In a model crystal with (0001), [101 0] initial crack, {101 1} first order pyramidal slip occurred at crack tip and following {101 2} twin was also observed. In the model crystal with (101 0)[0001] initial crack, the crack propagated parallel to initial crack plane with two prismatic slips in front of the crack. To understand strength of nano size titanium crystals as MEMS materials, stress intensity factor, K of each model crystals were also estimated from stress-strain curves obtained from these simulations.Copyright © 2005 by ASME