84 results about "Hybrid filter" patented technology

The invention discloses an intelligent photoelectric theodolite aerial target positioning and tracking system, which relates to the technical field of photoelectric detection, and comprises a multi-mode photoelectric sensor module integrating visible light, infrared thermal imaging, a laser radar and a polarized light sensor and supporting spectrum adaptive switching; the dynamic noise suppression processing module is used for eliminating environmental interference based on a time-space domain hybrid filtering algorithm; the multi-target tracking control module adopts a time-sharing partition scanning strategy and a graph neural network data association algorithm; the anti-interference servo driving module is used for realizing stable tracking under strong disturbance through inertial navigation-visual fusion compensation; and the edge computing platform is used for deploying a lightweight deep learning model to complete target recognition and trajectory prediction. According to the invention, through interdisciplinary collaboration of quantum dot materials, graph neural networks and physical equation constraints, the bottleneck of a single technology is broken through; and through closed-loop optimization of dynamic anti-interference and edge intelligence, full-link enhancement of'perception-decision-execution 'is realized.

The invention relates to the technical field of battery fault detection, and particularly discloses a new energy automobile battery fault intelligent diagnosis method and system, and the method comprises the steps: collecting the three-dimensional time sequence data flow of the voltage, temperature and internal resistance of a battery pack through a distributed sensor array; the environment temperature, the number of charge and discharge cycles and the vehicle operation condition parameters are synchronously integrated as auxiliary diagnosis dimensions, and a multi-source heterogeneous data sensing system is constructed; the method comprises the following steps: preprocessing original data by adopting a dual-channel hybrid filtering architecture based on a multi-source heterogeneous data sensing system, and establishing an adaptive filtering parameter adjustment mechanism to generate multi-modal data; according to the method, multi-dimensional time sequence data and auxiliary diagnosis parameters of the battery pack are comprehensively collected through a constructed multi-source heterogeneous data sensing system, a rich and accurate data basis is provided for fault diagnosis, noise interference is reduced through a two-channel hybrid filtering architecture and a self-adaptive filtering parameter adjustment mechanism, and the fault diagnosis accuracy is improved. And a solid foundation is laid for subsequent feature extraction and model training.

The invention relates to the field of intelligent warehousing, and discloses an intelligent clothing inventory real-time monitoring and automatic checking system based on RFID, which aims at solving the problems of poor environmental adaptability, high dynamic sensing delay, high cross-scene deployment cost and the like in the prior art, and comprises a reconfigurable RFID tag, an integrated spiral-dipole composite antenna and a temperature compensation circuit, dual-band switching of 860 to 960 MHz and self calibration are supported; a heterogeneous read-write network is formed by a fixed reader-writer and AGV mobile nodes, and signal coverage is enhanced through beam forming; the edge calculation module adopts an LSTM trajectory prediction and hybrid filtering algorithm to realize millisecond-level processing; the cloud platform maps the inventory in real time based on the digital twin model and traces the source of the block chain, the dynamic time slot distribution protocol optimizes the cooperation of the reader-writer, and the transfer learning model realizes cross-scene rapid adaptation. According to the invention, the tag reading rate is 99.7%, the inventory efficiency is improved by 300%, the deployment period is shortened by 80%, and the problems of accurate inventory management and dynamic tracking in a complex environment are solved.

The invention discloses an infrared CO2 high-precision detection system and method based on a self-adaptive balance signal, and belongs to the field of gas detection. The invention discloses an infrared CO2 high-precision detection system and method based on a self-adaptive balance signal. The infrared CO2 high-precision detection system comprises a sensing module, a signal conditioning and collecting module and a self-adaptive filtering processing module. According to the method, the problems of contradiction between response speed and stability, discrete signal tracking lag and sensitivity to environmental interference in the prior art are solved, the weight of a filtering channel is adaptively and dynamically adjusted, and the advantages of a fast filtering channel and a slow filtering channel are combined, so that balance between dynamic response and steady-state precision is realized; the problem of tracking lag of discrete signals is solved by analyzing the change rate of the signals in real time and dynamically adjusting a filtering strategy; and through a hybrid filter and an environment compensation algorithm, the influence of temperature and humidity changes on a detection result is reduced, so that the long-term stability of the sensor is improved.

The invention discloses a transformer voiceprint fault identification method and system. The method comprises the following steps: extracting a hybrid filter power spectrum feature vector from a target transformer voiceprint signal; according to a preset multi-granularity dynamic multi-dimensional feature fusion algorithm, fusing the hybrid filter power spectrum feature vector with a preset multi-dimensional feature vector to obtain an MDF vector of at least one frame of transformer voiceprint sub-signal, and splicing the MDF vectors of the frames of transformer voiceprint sub-signals to obtain an MDF vector of the transformer voiceprint sub-signal; generating a fusion feature vector corresponding to the target transformer voiceprint signal; constructing a transformer voiceprint recognition model based on a full-span sparse attention mechanism of a structured factor; and inputting the fusion feature vector into a transformer voiceprint recognition model, and outputting to obtain a fault recognition result. The problems of strong noise interference, weak feature discrimination, poor model robustness and the like in transformer fault voiceprint recognition in a high-noise environment are effectively solved, and the accuracy and robustness of fault recognition are remarkably improved.

The invention relates to the technical field of intelligent control and information security crossing, in particular to an anti-error data injection Kalman filtering method based on neural network enhancement, which comprises the following steps: firstly, constructing a neural network auxiliary correction module for identifying and correcting an extended Kalman filtering intermediate variable affected by an error data injection attack; secondly, designing a hybrid filtering architecture, and combining the learning ability of a neural network with the theoretical advantages of traditional Kalman filtering; and finally, realizing adaptive optimization of model parameters through a mixed training mechanism. Accurate state estimation can be achieved only through limited prior information, and meanwhile the calculation efficiency and interpretability of a traditional Kalman filter are kept. And the anti-interference capability of the system in a non-Gaussian noise environment is effectively improved. The application of the method on an induction motor model is obviously better than that of a traditional anti-error data injection scheme, and breakthrough progress is achieved in the aspects of estimation precision and robustness.

The invention provides a hybrid filtering denoising method and device based on adaptive parameters, and relates to the technical field of three-dimensional point cloud processing. The method comprises the following steps: acquiring three-dimensional point cloud data of a target object; performing voxel segmentation processing on the three-dimensional point cloud data to obtain a plurality of voxels; according to the segmented three-dimensional point cloud data, combining a plurality of voxels by using a density similarity model to obtain a voxel set with different density distributions; based on the merged voxels, a radius filtering parameter and a statistical filtering parameter of each voxel set are adaptively determined, the radius filtering parameter is used for filtering isolated noise in the three-dimensional point cloud data, and the statistical filtering parameter is used for filtering random noise in the three-dimensional point cloud data; performing random noise removal processing on the three-dimensional point cloud data according to the statistical filtering parameters; and performing isolated noise removal processing on the three-dimensional point cloud data after random noise removal according to the radius filtering parameter.

The invention relates to a tunable high-repetition-frequency high-power multi-wavelength optical pulse generation system, which belongs to the technical field of optoelectronic equipment and is structurally characterized in that the output end of a seed pulse source (1) is connected with the input end of a repetition-frequency multiplication module (2), and the output end of the repetition-frequency multiplication module (2) is connected with the input end of a shaping optical path module (3); the output end of the shaping light path module (3) is connected with the input end of the power amplification module (4), and the output end of the power amplification module (4) is connected with the input end of the pulse compression module (5); a multi-wavelength pulse seed source is designed by using a plurality of hybrid filters, and tunable multi-wavelength, high-repetition-frequency and high-power narrow pulse output is realized by using technologies such as an MZI-Sagnac hybrid repetition-frequency multiplication structure and the like.

To suppress degradation of characteristics, a high frequency module includes a mounting substrate, a first signal terminal, a second signal terminal, and ground terminals, and a hybrid filter. The hybrid filter is coupled between the first signal terminal and the second signal terminal. The hybrid filter includes an acoustic wave filter having at least one acoustic wave resonator, a first inductor, and a capacitor. The pass band width of the hybrid filter is larger than that of the acoustic wave resonator. A resin layer is disposed on a first principal surface of the mounting substrate. The resin layer covers at least part of the acoustic wave filter. A metal electrode layer covers at least part of the resin layer, and is coupled to the ground terminals. The second inductor is coupled between the hybrid filter and the second signal terminal.

The invention relates to the technical field of aerospace engineering, space robots and intelligent control, and particularly discloses a space debris intelligent capture and treatment method and system based on an SAC algorithm, and the method comprises the steps: obtaining the motion data of a space non-cooperative target and the system parameters of a space robot; constructing a hybrid filtering optimization framework based on the motion data and the system parameters; adjusting a hybrid weight and a noise forgetting factor in the hybrid filtering optimization framework by using an SAC algorithm to obtain self-adaptive estimation of a target attitude; a state space and an action space are defined based on a base of the space robot and a mechanical arm of the space robot respectively, and based on the state space and the action space, a mechanical arm movement path of the space robot is generated through an SAC algorithm in combination with a multi-target reward function; and the self-adaptive estimation of the target attitude is fed back to the motion path of the mechanical arm, dynamic cooperation of filtering and control is achieved, and stable capture of the space non-cooperative target is completed.

A bandpass filter is provided that includes series acoustic resonators and shunt acoustic resonators. The shunt acoustic resonators include a first transversely-excited film bulk acoustic resonators (XBAR) that includes a diaphragm comprising a portion of a piezoelectric layer that is over a cavity of the first XBAR and a first interdigital transducer (IDT) on a surface of the piezoelectric layer. The first XBAR is connected between (i) a node directly between a first series acoustic resonator of the plurality of series acoustic resonators and the input port, and (ii) a ground of the bandpass filter. The series and shunt acoustic resonators include one or more solidly-mounted XBARs that includes a portion of a second piezoelectric layer that is solidly mounted over a substrate with a Bragg reflector disposed therebetween.

The invention relates to the technical field of mobile telecommunication services such as new-generation mobile communication base station equipment, new-generation mobile communication core network and access network construction, networking and the like, in particular to a 6G communication and sensing integration method for multi-dimensional radio frequency characteristic intelligent analysis, which comprises the following steps of: S1, establishing a ground multi-antenna base station, and collecting radio frequency characteristics of a target; s2, establishing a self-adaptive path loss model, and realizing accurate estimation of a target distance by analyzing received signal strength data; s3, estimating an angle of arrival of the target signal in combination with an improved MUSIC-ESPRIT fusion algorithm based on the radio frequency characteristics; s4, based on the frequency change information of the target signal, estimating the distance position and the motion state of the target relative to the ground multi-antenna base station according to a Doppler frequency shift theory in combination with a hybrid filtering architecture driven by environmental perception; and S5, performing information fusion of different levels on the distance position and the motion state obtained by signal feature analysis to obtain a final target tracking result.

The invention belongs to the technical field of positioning and navigation, and discloses a precise single-point positioning method and device based on a single Beidou, and the method comprises the steps: building a three-frequency non-difference and non-combination precise single-point positioning observation model based on a Beidou satellite system, estimating the delay of an ionized layer through a non-difference and non-combination method, and extracting the electron content in a line-of-sight direction; carrying out iterative solution on the electron content in the sight line direction by using a multiplication algebraic reconstruction algorithm to obtain the electron density of the ionized layer; taking ionosphere electron density as prior information, and adding ionosphere constraint estimation to obtain a state vector of a receiver; and calculating a signal quality evaluation coefficient of the Beidou satellite system, and according to the signal quality evaluation coefficient, updating the state vector of the receiver by using a hybrid filter to obtain a final positioning position. According to the invention, high-precision regional ionosphere delay modeling and Beidou three-frequency non-difference and non-combination precision single-point positioning under ionosphere constraint are realized, and the positioning precision and robustness of a single Beidou precision single-point positioning system are effectively improved.

This invention discloses a computer vision-based method for ship traffic situational awareness, comprising: collecting ship traffic data and preprocessing it to obtain a video frame sequence; performing ship target detection and outputting a set of ship detection results; constructing an improved ByteTrack model and outputting a set of trajectory segments; performing a Poisson-Dobernoli hybrid filter to obtain a set of ship tracks; performing coordinate mapping and track element calculation, calculating speed, heading, and position; performing risk event determination and outputting risk event results; and performing graded early warning and interpretable processing to generate early warning information. This invention, through the collaborative processing of the improved ByteTrack model and the Poisson-Dobernoli hybrid filter, achieves graded early warning and interpretable output of risk events for ship targets under complex sea surface imaging conditions.

The invention discloses a lumped element and acoustic resonator mixed filtering balun and radio frequency front end system, which comprises a first module, a second module, a third module, a fourth module, a fifth module, a sixth module and a seventh module, and is characterized in that the first module, the third module and the fourth module are acoustic-electric mixed modules and comprise lumped elements and acoustic resonators; the first module, the fifth module, the sixth module and the seventh module are used for forming a plurality of transmission zero points outside a passband of the filtering balun so as to enhance the selectivity and out-of-band rejection of the edge of the passband, and the second module, the fifth module, the sixth module and the seventh module are electric modules, are formed by lumped elements and are used for realizing the passband of the required filtering balun. Reflection loss caused by impedance mismatching when discrete devices are cascaded can be avoided, insertion loss of a system is remarkably reduced, the size and cost of the system are reduced, the broadband advantage of an electrical filter and the high selectivity advantage of an acoustic filter are integrated, and meanwhile the high-performance filtering balun which is miniaturized, large in bandwidth and high in selectivity is achieved.

The invention provides an efficient image denoising method based on a hybrid filter technology, and belongs to the field of digital image processing. Traditional bilateral filtering and self-defined convolution filtering are fused to denoise an image; the method comprises the following steps: performing bilateral filtering on a noisy image for preprocessing, and performing self-defined convolution filtering on the preprocessed image to obtain a denoised image; the denoising precision can be improved while the real-time performance is ensured, and key pain points in the fields of industry, medical treatment, consumer electronics and the like are solved.

The invention provides a manufacturing method of a hybrid filter, and the method comprises the steps: providing a first substrate, and forming a piezoelectric layer on the first substrate; forming a first electrode and a bonding layer on the piezoelectric layer; an etching groove and a connecting conductive column are formed in the bonding layer, the first electrode at the preset position is exposed in the etching groove, and the connecting conductive column penetrates through the bonding layer and is connected with the piezoelectric layer and / or the first electrode; a second substrate is provided, a passive element and a leading-out conductive column are arranged in the second substrate, the leading-out conductive column is electrically connected with the passive element, the bonding layer and the second substrate are bonded, and the leading-out conductive column is electrically connected with the connecting conductive column; the first substrate is removed, and a second electrode is formed on the side, away from the second substrate, of the piezoelectric layer. According to the manufacturing method of the hybrid filter, the passive element is arranged in the second substrate, integrated connection of the passive element is achieved in the bonding process of the bonding layer and the second substrate, the process is simple, the device size can be reduced, and the cost can be reduced.

The invention discloses a hybrid filter, and relates to the technical field of radio frequency. The hybrid filter comprises a substrate, a piezoelectric layer, a transmission line resonator, an acoustic wave device and a connecting layer, wherein the piezoelectric layer is arranged on the surface of the substrate, and the piezoelectric layer is made of a high-dielectric-constant piezoelectric material; the transmission line resonator is arranged on the surface of the piezoelectric layer or arranged in the piezoelectric layer; the acoustic wave device is arranged in the piezoelectric layer; and the connecting layer is arranged on the surface of the piezoelectric layer and is used for connecting the transmission line resonator and the acoustic wave device. Therefore, the transmission line resonator based on the high-dielectric-constant piezoelectric layer and the acoustic wave device in the same piezoelectric layer form a three-dimensional and layer-stacked hybrid filter structure, so that the miniaturization of the hybrid filter formed by the high-dielectric-constant transmission line resonator and the acoustic wave device is realized; therefore, the problems of impedance mismatch, insertion loss superposition and the like are effectively reduced while the excellent performances of high selectivity, wide band and high power tolerance are realized.

Aspects of the present disclosure include a filtration system. The filtration system includes a mask and a hybrid filter for use with an activated carbon-based filter and attachable to the mask. The mask may include an interface around a periphery of the mask for sealing engagement with a skin surface of a user. The hybrid filter may include a chamber configured to receive the activated carbon-based, an electro-ionic filter, and a Faraday cage. The electro-ionic filter is in series with the material-based filtration element and includes a plurality of electrodes arranged in series with each other. The electrodes of the plurality of electrodes are evenly spaced apart from each other in an alternating positive-polarity negative-polarity sequence. The Faraday cage surrounds at least the electro-ionic filter.

This invention provides a performance verification device and method for a ground-to-air frequency domain electromagnetic detection receiving system, belonging to the field of ground-to-air detection technology. It includes a transmission link for generating a standardized single-frequency excitation magnetic field, comprising an inverter and modulation module, a hybrid filtering module, a dynamic power supply amplifier module, and a transmission calibration coil module connected in sequence; and a receiving and processing link for receiving the excitation magnetic field and evaluating coil performance, comprising a receiving coil module, an attitude correction and noise suppression module, a data processing module, and a performance evaluation module. This invention enables rapid verification of the receiving device, especially in high-altitude areas, allowing for the early detection of potential equipment problems through automated processes, reducing time and economic losses caused by equipment failure.

In the high-frequency module (100) of the present invention, the hybrid filter includes an elastic wave filter (10) having at least one elastic wave resonator, an inductor (L11) having a winding portion (413), and a capacitor. The elastic wave filter has a plurality of external electrodes including a first input / output electrode (15) connected to a first signal terminal, a second input / output electrode (16) connected to a second signal terminal, and a ground electrode (17) connected to a ground terminal. The inductor (L11) is arranged on a first main surface (41) of a mounting substrate (4) and is adjacent to the elastic wave filter (10) when viewed from above in the thickness direction of the mounting substrate. When viewed from the direction of the winding axis (A4) of the winding portion (413) of the inductor, the inner portion (416) of the winding portion of the inductor does not overlap with any of the first input / output electrode, the second input / output electrode, and the ground electrode.

The invention belongs to the technical field of vibration precision test, and discloses a data acquisition and processing method for satellite micro-vibration ground test. Comprising five core steps of original vibration signal acquisition, adaptive oversampling, analog-digital hybrid filtering, dynamic extraction recovery and micro-vibration transfer characteristic extraction. Through an analog-to-digital hybrid filtering architecture and in combination with an intelligent oversampling-extraction noise reduction technology, micro-vibration signal measurement with a high signal-to-noise ratio is realized. The method has the main technical advantages that various test items can be completed by using a single-specification sensor; a dynamic parameter adjustment mechanism can automatically optimize a processing strategy according to signal characteristics, and the test adaptability is remarkably improved.

The utility model discloses an analog-digital hybrid filter for electrocardiosignal noise reduction processing, and relates to the field of signal noise reduction. The analog filter is a band-pass filter and comprises a low-pass filter and a high-pass filter which are connected in sequence, and the low-pass filter comprises a series inductor and a parallel capacitor; the high-pass filter comprises a parallel inductor, a series capacitor and a radio frequency port connected with the low-pass filter, and the series capacitor in the high-pass filter and the series inductor in the low-pass filter are connected in series to form radio frequency signal output; the output of the radio frequency signal is connected with the signal amplifier, the output port of the amplifier is connected with the analog signal input end of the ADC converter, and the digital signal output port of the ADC converter is divided into a plurality of paths which are respectively connected to the digital filter. By combining analog filtering and digital signal processing technologies, the quality of electrocardiosignals can be effectively improved, the real-time performance of signal processing is improved, and the design flexibility is enhanced.

The invention provides a cloth defect detection method and device based on machine vision, and relates to the technical field of cloth defect detection methods, and the method comprises the steps: collecting a gray level image of a cloth surface, and generating a mixed filtering core to filter pixel points of the image; constructing a double-flow feature extraction model to extract a microcosmic feature map and a texture feature map, constructing a channel attention weight coefficient by adopting channel splicing, generating a space attention weight coefficient of each pixel point at the same time, and performing multi-scale feature fusion on the channel attention weight coefficient and the space attention coefficient of the pixel point to obtain a multi-scale feature fusion image; generating a fused feature map; and constructing a thermodynamic diagram of the feature map by using a convolutional network, constructing an adaptive threshold, and generating a binarized image of the flaws. According to the design of the double-flow feature extraction model, the characterization of multiple types of flaws is improved, a weight coefficient is generated through channel splicing, an important feature channel is amplified, a suspected flaw area is focused, and the thermodynamic statistical characteristics of a local area are embodied through the use of a thermodynamic diagram.

A glass substrate-based hybrid filter structure includes a primary die and secondary and tertiary dies stacked on the sides of the primary die, the primary die including a glass substrate, a glass semiconductor device disposed on the glass substrate, and a glass via disposed within the glass substrate, the secondary die including a surface acoustic wave filter (SAW), and the tertiary die including a bulk acoustic wave filter (BAW), the SAW and the BAW being electrically connected to the glass semiconductor device by bumps. By configuring two acoustic filters into a single filter, the acoustic filter on one surface provides steep rejection at one frequency, while the acoustic filter on the other surface provides rejection at a different frequency, achieving a sharp filter roll-off and increased bandwidth, while the glass substrate provides superior material product reliability over silicon substrates and is suitable for high frequency applications, resulting in superior electrical performance.

The invention relates to the field of intelligent traffic and vehicle-mounted automatic control, in particular to a taxi automatic top light control system based on an intelligent camera, which comprises the following steps of: acquiring video stream data in a taxi in real time through the camera, and analyzing the video stream in real time by adopting a double-branch heterogeneous network structure based on a deep neural network model; carrying out multi-frame fusion filtering processing on an analysis result, and generating a passenger existence confidence coefficient curve by adopting a hybrid filtering algorithm; decision making is carried out according to the confidence coefficient curve, a dual protection mechanism is set, and when the confidence coefficient curve continuously exceeds a first preset threshold value and reaches a second preset time threshold value, a ceiling lamp state switching instruction is triggered; an instruction from the decision control module is received through a vehicle-mounted communication bus, and automatic switching of the state of the ceiling lamp is achieved in a differential signal transmission mode. The video stream is analyzed from different angles based on the deep neural network, and the accuracy of passenger state analysis is improved.

The invention discloses a non-cooperative target searching method based on identification-planning joint optimization, and belongs to the technical field of unmanned aerial vehicles. A terrain perception multi-mode hybrid filter is constructed to accurately predict a trajectory after a target is separated from a visual field, high-reliability re-identification of the target is realized in combination with a fine-grained feature decoupling and dynamic fusion mechanism, and an identification-planning joint optimization adaptive search method is adopted to generate an optimal search path. The problems of long-time tracking and re-identification of the non-cooperative target in a complex environment are solved, and the continuous working capability of an unmanned aerial vehicle monitoring system is remarkably improved.

The invention relates to the technical field of vehicle safety control, in particular to an air bag collision decision-making method and device, computer equipment, a storage medium and a computer program product, and is implemented based on an air bag collision decision-making processor, and a self-adaptive hybrid filter module and a high-precision hardware integrator module are integrated in the processor. The method comprises the following steps: acquiring acceleration sensor data of a vehicle in parallel based on a pre-deployed sensor interface, and inputting the acceleration sensor data into a self-adaptive hybrid filter module for processing to obtain acceleration sensor data after filtering processing; inputting the filtered acceleration sensor data into a high-precision hardware integrator module, and carrying out parallel calculation on a speed value and a displacement value in a preset clock period; and analyzing the speed value and the displacement value based on a preset collision judgment rule, and executing an airbag control decision based on an analysis result. By adopting the method, the real-time performance of airbag decision can be improved and the CPU load can be reduced on the premise of ensuring the collision detection accuracy.

The invention discloses an efficient hybrid filter cabinet and a dynamic harmonic suppression method thereof.According to the filter cabinet, the power factor of a power grid is detected in real time through a three-phase reactive compensation capacitor, when the power factor is lower than a set value, the capacitor automatically compensates reactive power, the transmission efficiency of the power grid is improved, and line loss is reduced; the three-phase harmonic filter reactor is used in cooperation with the capacitor, load harmonic waves in a power grid are effectively eliminated, equipment is protected from being damaged by the harmonic waves, the service life is prolonged, the state of the power grid is dynamically monitored through the control module, the harmonic problems are accurately judged, the working states of the capacitor and the reactor are adjusted according to the monitoring result, and real-time harmonic suppression is achieved. The control module also endows the filtering cabinet with an intelligent management function, and through remote monitoring and data analysis, managers can know the operation state of a power grid in real time, find and process potential problems in time, reduce the operation and maintenance cost, and improve the management efficiency.

This invention provides a machine vision-based method and apparatus for detecting fabric defects, belonging to the technical field of fabric defect detection methods. The method includes acquiring a grayscale image of the fabric surface and generating a hybrid filter kernel to filter the pixels of the image; constructing a two-stream feature extraction model to extract microscopic and texture feature maps; using channel concatenation to construct channel attention weight coefficients and simultaneously generating spatial attention weight coefficients for each pixel; fusing the channel attention weight coefficients and the spatial attention weight coefficients of the pixels at multiple scales to generate a fused feature map; using a convolutional network to construct a heatmap of the feature map; constructing an adaptive threshold to generate a binarized image of the defect. The two-stream feature extraction model design improves the characterization of multiple types of defects. By generating weight coefficients through channel concatenation, important feature channels are magnified, focusing on suspected defect areas. The use of the heatmap reflects the thermal statistical characteristics of local areas.