36 results about "Peak detection algorithm" patented technology

The invention relates to the technical field of image recognition, in particular to a plate defect recognition and detection method based on an image recognition model. The method comprises the following steps: aiming at the technical problems of insufficient texture direction identification precision and difficulty in defect type distinguishing in a traditional plate surface detection method, the invention provides an intelligent detection solution based on polarization imaging. A multi-angle polarization measurement system is constructed through a fixed polarizer and a rotatable analyzer, and omnibearing polarization characteristic collection of the surface of a steel plate is achieved. A polarization intensity change model is established based on the Malus law, the polarization degree and polarization angle parameters of each pixel point are extracted, and a polarization parameter spatial distribution diagram reflecting the surface microstructure is constructed; through statistical analysis of polarization angle distribution characteristics, a rolling direction dominant angle is identified by adopting a peak detection algorithm, and accurate classification and severity evaluation of defects are realized in combination with anisotropy characteristic analysis.

The invention provides an unmanned aerial vehicle nest site selection planning method and system and a related device, and relates to the technical field of unmanned aerial vehicles, and the method comprises the steps: converting the latitude and longitude of a tower and the altitude into a 3D voxel grid; extracting spatial features of a tower point group from the 3D voxel grid by using a 3D convolutional neural network; a density peak detection algorithm is applied to the spatial features to determine a clustering center, a decision value is calculated according to the local density and the minimum distance between the local density and the high-density point, and the optimal clustering point is dynamically identified based on the decision value to serve as initial site selection point distribution; and on the basis of the initial site selection point distribution, optimizing the site selection point distribution by using a reinforcement learning model. According to the technical scheme of the invention, the problems of time and labor consumption and difficulty in processing large-scale tower point group data in manual site selection in the prior art can be effectively solved; a traditional method is difficult to comprehensively consider multi-dimensional factors such as signal coverage, economic cost and geographical constraints; and a site selection strategy cannot be dynamically adjusted, and complex geographical environments and task requirements are difficult to deal with.

The invention provides a self-decoupling method for multi-characteristic spectrum inversion and a ribbon-parallel fiber grating sensing optical cable, and relates to the technical field of optical fiber sensing, and it needs to be explained that a grating reflection spectrum of the ribbon-parallel fiber grating sensing optical cable is obtained by determining a reference quantity according to the embodiment of each method; the method comprises the following steps: placing a ribbon-parallel fiber bragg grating sensing optical cable in a preset characteristic calibration state, collecting a reflection spectrum of the ribbon-parallel fiber bragg grating sensing optical cable, and traversing the reflection spectrum through a main peak global extremum search algorithm to obtain a central wavelength of the ribbon-parallel fiber bragg grating sensing optical cable; extracting a first-order left side lobe intensity and a first-order right side lobe intensity in symmetrical wavelength windows at two sides of the central wavelength by adopting a peak detection algorithm, and respectively calculating a wavelength reference variation, a left side lobe reference variation and a right side lobe reference variation according to the calibration reference quantity; and constructing a response characteristic calibration model so as to respectively solve the temperature variation and the strain variation of the ribbon fiber grating sensing optical cable. According to the invention, the decoupling precision and stability can be improved.

The invention provides a water quality suspended matter detection system and method based on sensing data, and belongs to the technical field of water quality suspended matter detection. Gain compensation is performed on a mixed optical signal based on backward spontaneous radiation noise power, so that the influence of self gain drift of a sensor and environmental interference on a signal baseline is effectively eliminated; the adaptive variational mode decomposition algorithm decomposes the standardized signal into a plurality of intrinsic mode functions, accurately extracts signals strongly related to the physical characteristics of the target suspended particulate matter, generates a particulate matter characteristic waveform signal with an optimized signal-to-noise ratio through the intrinsic mode functions screened out through weighted stacking, further highlights the time-frequency characteristics of the target signal, and improves the accuracy of the particulate matter characteristic waveform signal. Each independent particulate matter signal event is accurately separated by adopting a dynamic threshold peak value detection algorithm, and the detection accuracy is improved by extracting a multi-dimensional feature vector from each signal event and distinguishing target particulate matters from interferents by utilizing a pre-trained event classifier.

The invention provides a power distribution network project progress regulation and control method, system and device and a storage medium, and relates to the technical field of power distribution network management, and the method comprises the steps: obtaining project progress related data, including measure input data and progress change data; performing similarity matching on the progress change data and a theoretical progress mode cluster in a preset theoretical cluster library to obtain a matched progress cluster, and calculating progress reference features; performing feature extraction on the input data of the single measure to obtain a project basic feature, an environment interference feature and a measure core feature; calculating an instantaneous gain value sequence of the single measure in the observation time window by utilizing a time sequence gradient lifting tree model and combining the characteristics; determining the cumulative progress contribution degree of the single measure through a peak detection algorithm; generating an adjustable vector of the single measure in the to-be-executed process; and in combination with the confidence interval parameter of the matched progress cluster and the project target progress, generating a regulation and control scheme and issuing the regulation and control scheme to an execution end. According to the invention, the precision of progress regulation and control is improved.

The invention relates to a method and device for detecting the liquid level position in an oil casing annulus. The method comprises the following steps: acquiring a temperature distribution data sequence of an oil well acquired by a distributed optical fiber sensing system; feature points caused by the tubing coupling are recognized through a peak detection algorithm, and the number and positions of the feature points are recorded; obtaining a first liquid level according to the number of the feature points; determining a second liquid level based on local temperature characteristics of the temperature distribution data sequence; and calculating the liquid level depth according to the feature point position, the first liquid level and the second liquid level. By the adoption of the method, continuous and real-time working fluid level detection can be achieved through existing distributed optical fiber sensing data, dependence on special hardware is avoided, and system complexity and cost are remarkably reduced on the basis that measurement robustness and accuracy are improved.

The present application relates to a kind of lightning intensity prediction method and system based on deep time sequence learning, belong to lightning early warning technical field, solve the problem of insufficient fine prediction ability of lightning intensity in prior art, peak time positioning is not accurate.The method comprises: obtaining current atmospheric electric field intensity data, using peak detection algorithm, determine the current peak position;According to the current atmospheric electric field intensity data, determine the current intensity time sequence data;Current peak position, and current intensity time sequence data, input lightning intensity prediction model, determine the first predicted peak position and the first predicted intensity peak;Wherein, lightning intensity prediction model includes time feature fusion layer, encoder layer and double-head output layer.The accurate timing and quantitative prediction of sparse, sudden lightning intensity peak are realized.

This invention provides a method, system, equipment, and storage medium for project schedule control in power distribution networks, relating to the field of power distribution network management technology. The method includes: acquiring project schedule-related data, including data on measure inputs and schedule changes; performing similarity matching between the schedule change data and theoretical schedule pattern clusters in a pre-set theoretical cluster library to obtain matched schedule clusters, and calculating schedule baseline features; extracting features from the input data of individual measures to obtain basic project features, environmental interference features, and core feature features of the measures; using a time-series gradient boosting tree model, combined with these features, calculating the instantaneous gain value sequence of individual measures within the observation time window; determining the cumulative schedule contribution of individual measures through a peak detection algorithm; and generating an adjustable vector for the individual measures in the pending execution process based on this; and generating a control plan and distributing it to the execution end by combining the confidence interval parameters of the matched schedule clusters and the project target schedule. This invention improves the accuracy of schedule control.

The invention relates to the technical field of frequency measurement, and discloses a short-wave carrier center frequency measurement method, which comprises the following steps: collecting short-wave signal data, and carrying out parallel processing to generate multi-domain feature representation; constructing a multi-scale time-frequency analysis framework by applying fractional order wavelet transform; fractal dimensions and self-similarity indexes of the signals are calculated, and geometric features are extracted; combining fractal features and noise type recognition results to optimize fractional order parameters; feature fusion is executed in a multi-domain multi-scale space, and enhanced frequency representation is constructed; accurately positioning the carrier center frequency by applying a hierarchical peak detection algorithm; according to the method, the accuracy and the reliability in the short-wave carrier are remarkably improved while the complexity of the algorithm is kept controllable, and particularly, the method has obvious advantages in a low signal-to-noise ratio and complex interference environment, so that powerful support is provided for improving the performance of a short-wave communication system.

The invention relates to a resonance protection method and device, equipment, a storage medium and a program product. The method comprises the following steps: acquiring voltage and current data of a target node of a to-be-tested power system; determining an inter-harmonic set and an inter-harmonic subset from the voltage and current data; if the root-mean-square values of the inter-harmonic set and the inter-harmonic subset meet the triggering condition of resonance protection, extracting peak and valley moments of voltage from the voltage and current data by using an adaptive multi-scale peak detection algorithm; determining resonance protection parameters of the target node according to the peak and valley moments; and performing resonance protection on the target node according to the resonance protection parameter. By adopting the method, the anti-interference capability can be improved.

The present application relates to the field of virtual reality, in particular to a lung rehabilitation nursing psychological counseling method and system based on virtual reality. The method collects the chest and abdominal girth change signal of the patient, calculates the respiratory flow rate curve through band pass filtering and peak detection algorithm, identifies the inspiration starting point, expiration starting point and breath holding starting point on the respiratory flow rate curve as the respiratory phase turning point. The digital phase-locked loop mechanism is introduced, the respiratory phase turning point is taken as the reference clock signal, the actual presentation time of the virtual visual and auditory stimulation is taken as the output signal of the voltage-controlled oscillator, the rendering pipeline delay of the visual and auditory generation component is adjusted through closed-loop feedback, so that the phase delay of the virtual visual and auditory stimulation is constantly within the preset neuromuscular response threshold. The present application eliminates the phase lag and drift phenomenon between the audio-visual guide signal and the real breathing action of the patient, and enhances the neural drive synchronization of the diaphragm and auxiliary respiratory muscles under the rhythmic audio-visual stimulation.

The present application belongs to the field of frequency hopping network station parameter estimation, and particularly relates to a frequency hopping period estimation method based on FFT transformation. In this scenario, mainly containing frequency hopping network station signals and other interference signals, the characteristics of conventional ultrashort wave frequency hopping network station fixed frequency hopping period can be utilized, the time-frequency peak value of frequency hopping pulse is obtained through a time-frequency peak value detection algorithm, the time center sequence of frequency hopping pulse is created according to the time-frequency peak value, and the FFT transformation processing is performed on the time center sequence, so that the peak value coordinate information related to the frequency hopping period can be read in the transformed FFT spectrum, and the value of the frequency hopping period can be calculated according to the information coordinate. The present application can obtain accurate estimation value under the condition of poor detection (more missed detection frequency hopping numbers) and low time resolution of time-frequency diagram, and the method is simple and has good effect.

The invention relates to the technical field of laser-induced breakdown spectroscopy, in particular to an LIBS (laser-induced breakdown spectroscopy) technology-based real-time processing and quantitative analysis system for spectral data, which comprises a laser emission module, a spectrum acquisition module, a real-time processing unit, a quantitative analysis module, an environmental parameter monitoring module and a remote communication module, the sub-second response speed is realized through an FPGA parallel processing architecture, the signal processing precision is improved by adopting an improved wavelet threshold algorithm and a self-adaptive peak position detection algorithm, and a multi-element analysis error is controlled to be lt through a matrix effect correction algorithm; 5%. The system is provided with an environmental parameter monitoring module and a remote communication module, achieves environmental interference suppression and remote data transmission, is suitable for rapid element quantitative analysis scenes in industries such as metallurgy, environmental protection and geological exploration, and has the remarkable advantages of being high in real-time processing capacity, high in quantitative precision, high in anti-interference capacity and the like.

Systems / techniques are provided for facilitating post-detection chromatographic peak validation. In various embodiments, a system can cause a chromatography device coupled to a mass spectrometer to scan a specimen, thereby yielding a chromatogram and mass spectra. In various aspects, the system can identify, via a peak detection algorithm, a plurality of purported peaks in the chromatogram. In various instances, the system can separate, via execution of a machine learning classifier on respective ones of the plurality of purported peaks, the plurality of purported peaks into a set of valid peaks and a set of invalid peaks. In various cases, the system can perform a mass spectrometry analysis on first portions of the mass spectra that correspond to the set of valid peaks but not on second portions of the mass spectra that correspond to the set of invalid peaks.

The invention discloses a bionic fish tail swing non-contact measurement method, device and equipment based on machine vision, a medium and a product, and relates to the technical field of bionic robots and hydrodynamic force testing. The method comprises the following steps: acquiring a video stream containing motion of a target bionic fish; performing bionic fish target area positioning and image processing on each video frame in the video stream to determine fish body positioning information of a target bionic fish; the fish body positioning information comprises fish body reference point information and fishtail end point information; determining fishtail swing measurement and analysis information according to the fish body positioning information by adopting an improved peak detection algorithm and a fluid mechanics model; the fishtail swing measurement and analysis information comprises average swing period, frequency, swing force and output power; wherein the fluid mechanics model is a mathematical model which is determined by adopting a Morison equation on the basis of fluid mechanics. According to the device and the method, the swinging condition of the fishtail of the bionic fish can be accurately analyzed and measured in a non-contact manner.

A method to detect a bearing fault in a rotating machine is provided. The method includes receiving current data from the rotating machine during operation at a sampling frequency, obtaining a current frequency spectrum for a sample of the received current data, reducing noise in the current frequency spectrum to create a reduced current frequency spectrum, performing a peak detection algorithm to detect peak amplitude values in the reduced frequency spectrum for a frequency range of the current frequency spectrum, determining a bearing fault from the peak amplitude values in the reduced current frequency spectrum, and responsive to the determination of the bearing fault, generating an alarm.

The present application relates to a Laos language unsupervised phoneme segmentation method based on multi-feature interaction fusion, and belongs to the technical field of natural language processing. The present application comprises the following steps: collecting and processing Laos language audio data sets; extracting self-supervised features, spectral features and pitch features; fusing the self-supervised features, spectral features and pitch features to train a phoneme segmentation model; based on the principle of probability contrast loss, distinguishing adjacent frames and random interference frames to realize optimization of the phoneme segmentation model; in the inference stage, inputting the phoneme segmentation model output vector into a peak detection algorithm to generate the final phoneme boundary. The present application can better interact and fuse the advantages of different features, overcome the limitations of the direct addition method, and thus achieve better results in overall performance.

The application discloses a kind of underwater rotating machinery multi-component vibration signal decoupling methods based on order analysis.Method includes: first, the discrete time domain vibration signal of underwater rotating machinery is obtained by superimposing multiple harmonic components;Then the discrete time domain vibration signal is transformed into angular domain signal, performs discrete Fourier transform, and extracts the amplitude spectrum of each order component;On this basis, the order amplitude spectrum is identified by using multi-peak detection algorithm, and the main order component order set is screened out;Finally, the time domain waveform corresponding to the main order component is extracted by order decoupling method.The application can effectively decouple the multi-component vibration signal in underwater rotating machinery, and provide quantitative basis for locating excitation source, analyzing excitation source characteristics or fault diagnosis.

This invention discloses a method and system for early warning of arrhythmias in smartwatches based on deep learning. The method first acquires PPG signals using a photoplethysmography (PPG) sensor and performs noise reduction and standardization using an adaptive filtering algorithm. Then, it uses a wavelet transform peak detection algorithm to identify heartbeat locations and obtain a heart rate variability (HRV) feature sequence. Next, it uses an improved Gramian angular field transform algorithm to convert the one-dimensional HRV feature sequence into a two-dimensional heart rhythm image. Then, it uses a ResNet-50 convolutional neural network to train an arrhythmia classification model, achieving automatic identification of different types of arrhythmias. Finally, it uses sliding window technology and an early warning mechanism to achieve continuous monitoring and timely alarms. This invention effectively solves key technical problems in existing arrhythmia detection technologies, such as low detection accuracy, poor real-time performance, high equipment costs, and unsatisfactory user experience, providing an innovative solution for heart rate health monitoring in smart wearable devices.

The invention discloses a Young modulus measuring method and a Young modulus measuring system for experiment teaching. An optical magnification lens group and a camera are sequentially arranged in the front edge optical axis direction of a fixed column at the bottom end of a test sample, an image of the fixed column is collected, and the positions of scale lines of a reticle and mark lines on the fixed column are recognized through a horizontal projection peak detection algorithm and an image processing algorithm of a multi-line equidistant recognition algorithm. High-precision automatic measurement of shrinkage and elongation of the test sample is realized; the AI tutoring module automatically judges whether an operation error exists or not according to the acknowledged value; the problems of low measurement precision, insufficient guidance, low efficiency and the like in a traditional physical experiment are solved, and the experiment teaching quality is remarkably improved; the device is suitable for experiment teaching of Young modulus measurement in university physical experiments, and can be expanded and applied to industrial or scientific research scenes of material mechanical property testing.

This invention discloses a method and system for quantitative identification of T / C lines on test strips based on QR codes. First, the image is preprocessed to achieve high QR code recognition rate under complex lighting and shooting angles. Then, the test strip is externally corrected using the QR code, and the approximate location of each color area is estimated based on the relative position of the QR code and the color development area. Next, the approximate location of the color area is further precisely estimated using the connected component method, and the image is cropped into sub-images. These sub-images are then subjected to horizontal and vertical grayscale projection, and peak detection algorithms are used to locate the boundaries of the color development areas and the positions of the C and T lines. Finally, the grayscale ratio of the T and C lines is calculated based on their average grayscale values. Furthermore, the C and T lines are drawn onto rectangular sub-images, and the sub-images are reverse-corrected and pasted back onto the original image. This image and the T / C line ratio are output to a mobile terminal, facilitating verification by mobile terminal users and improving the reliability of the results.

The application provides a water quality suspended matter detection system and method based on sensing data, and belongs to the technical field of water quality suspended matter detection.The gain of a mixed optical signal is compensated based on backward spontaneous radiation noise power, so that the influence of sensor gain drift and environmental interference on the signal baseline is effectively eliminated, an adaptive variational mode decomposition algorithm is used to decompose the normalized signal into multiple intrinsic mode functions, the signal strongly related to the physical characteristics of the target suspended particulate matter is accurately extracted, the intrinsic mode functions selected through weighting and superposition are used to generate a particulate matter characteristic waveform signal with optimized signal-to-noise ratio, the time-frequency characteristics of the target signal are further highlighted, a dynamic threshold peak detection algorithm is used to accurately separate each independent particulate matter signal event, multi-dimensional feature vectors are extracted from each signal event, and a pre-trained event classifier is used to distinguish the target particulate matter from interference, so that the detection accuracy is improved.

The invention discloses a self-adaptive parameter adjustment method based on prior information and a one-dimensional range profile peak detection algorithm, and the method comprises the steps: obtaining multi-band ISAR echo signals, and carrying out the splicing of the ISAR echo signals to form sparse aperture echo signals; constructing an optimization objective function and initializing related parameters, wherein the related parameters comprise regularization parameters; the regularization parameter is used as the input of an ADMM algorithm, the sparse aperture echo signal is reconstructed, and a multi-band fusion ISAR echo is obtained; peak value detection is carried out on the one-dimensional range profile of the multi-band fusion ISAR echo, and the number of detected peak values is evaluated according to the number of known targets in prior information; and according to an evaluation result, self-adaptively adjusting related parameters until the reconstructed multi-band fused ISAR echo meets a preset requirement, and obtaining an optimal regularization parameter. The method can be well compatible with various parameters, needing to be manually adjusted, of the ADMM algorithm, and the automation degree of parameter adjustment is improved in various scenes.

The application relates to an underwater seismic source monitoring device of an internal seven-direction arrangement silicon microphone, which comprises an internal seven-direction silicon microphone device, a multi-channel acquisition system, an underwater seismic source simulation system and a data processing system; the internal seven-direction silicon microphone device comprises an internal seven-direction support and seven silicon microphones fixed on the internal seven-direction support; a recess is arranged on the inner side wall of the internal seven-direction support, the recess comprises a hexagonal bottom surface and six inclined surfaces arranged around the hexagonal bottom surface, one silicon microphone is fixed on the hexagonal bottom surface, and one silicon microphone is fixed on each of the six inclined surfaces. The seven silicon microphones are arranged in a specific arrangement mode, can be conveniently installed on underwater equipment to collect sound signals, the collected sound signals can be used to accurately calculate specific coordinate points of a simulation seismic source and the main wave transmission direction of the simulation seismic source, and a peak value detection algorithm is adopted to analyze the distribution and evolution law of different events in the underwater seismic source simulation experiment process.

The invention discloses a coal mine underground RS485 two-parameter self-adaptive method, which comprises the following steps of: S1, multiplexing a serial port receiving pin of a single chip microcomputer into a timer input capture mode, and capturing original pulse width data of an RS485 bus; s2, processing low-level pulse width data in the original pulse width data through an improved multi-stage peak detection algorithm, and calculating to obtain the narrowest pulse width; s3, calculating the Baud rate according to the narrowest pulse width, and verifying the effectiveness of the Baud rate; s4, converting the original pulse width data into a digital sequence, and locking a correct check mode based on check bit identification of the digital sequence; and S5, performing data validity verification on the application layer, if the verification is passed, ending the two-parameter self-adaption, and if the verification is not passed, returning to the step S1. According to the coal mine underground RS485 two-parameter self-adaption method, two-parameter self-adaption matching is carried out from the Baud rate and the verification mode of the equipment, and the method has the advantages of being high in matching speed, high in precision, high in interference resistance, good in adaptability and high in practicability.

The invention relates to a power supply voltage dynamic adjustment method and circuit and a D-type power amplifier, and the method comprises the steps: carrying out the N-time up-sampling of an input signal, and obtaining a sampling result; wherein N is an integer greater than 1; carrying out amplitude detection on the sampling result by adopting an amplitude detection algorithm to obtain an amplitude detection result; the amplitude detection algorithm is used for carrying out amplitude detection on a sampling result by adopting a peak detection algorithm combining a translation window and a sliding window; determining a dynamic power supply voltage control signal according to an amplitude detection result by adopting a mode of combining multi-gear threshold comparison and a lookup table; the power supply voltage of the power supply is adjusted according to the dynamic power supply voltage control signal, and the power supply voltage is used for supplying power to an amplification output stage circuit of the D-type power amplifier. According to the method, the efficiency of the D-type power amplifier can be improved, the loss is reduced, and the overall system efficiency of the power amplifier and the power supply is improved in the application scene of battery power supply.

The invention provides a fracture test acoustic emission and high-speed imaging synchronous monitoring system and monitoring method, and belongs to the technical field of material fracture test monitoring. The monitoring system comprises a loading test device, an acoustic emission monitoring device, a high-speed imaging device and a synchronous control system, the synchronous control system collects a load signal output by the loading sensor and generates a synchronous pulse through a peak detection algorithm when the load reaches a peak value, and the synchronous pulse is recorded by the acoustic emission system and triggers the high-speed camera at the same time, so that a unified time reference is established, and accurate alignment of acoustic emission data and high-speed image data is realized. The system is simple in structure, high in synchronization precision and capable of stably capturing the rapid crack initiation and expansion process near the fracture peak value.

The application discloses a reconstruction method, in particular to a ground digital television multimedia broadcast reference signal high-quality reconstruction method, which comprises the following steps: step 1, high-precision synchronization and parameter tracking; step 2, robust channel estimation and equalization; step 3, soft information demapping and iterative decoding; step 4, bit stream re-modulation. The two-stage synchronization mechanism combining coarse synchronization and fine synchronization, combined with the improved windowing peak detection algorithm and dynamic threshold, can effectively resist noise and multipath interference, and significantly reduce the false alarm and missed detection probability. The joint tracking loop continuously compensates the carrier frequency deviation and the sampling clock deviation, effectively suppresses the phase noise and the inter-subcarrier interference, and ensures stable timing and carrier synchronization under high-speed movement and severe channel conditions.

The invention relates to the technical field of partial discharge, in particular to a multi-source partial discharge mode recognition method and system, and the method comprises the steps: firstly constructing a target discharge mode through a defect simulation module, then starting a voltage regulating table, gradually increasing the output voltage of a test transformer to a discharge critical value, and triggering partial discharge; an optical signal generated by discharging is captured by the fluorescent optical fiber in the corresponding area and is guided into the PMT through the transmission optical fiber; the PMT converts an optical signal into an electric signal under the power supply of the switching power supply, and transmits the electric signal to the oscilloscope in real time; the oscilloscope stores the electric signal data periodically and screens effective discharge signals through a peak detection algorithm to complete data cleaning; and inputting the training set into an optimized two-stage model of decision tree coarse classification and XGBoost fine classification, completing training, and realizing efficient and accurate identification of partial discharge types.

The invention discloses a deep learning-based X-ray image lossless counting method and device for cartons, and the method comprises the following steps: obtaining gray images of different energy levels, generating a pseudo-color image, optimizing the exposure time and dosage according to the material and thickness of a carton body, and carrying out the projection collection of four angles at the same time; the collected images are preprocessed; inputting the preprocessed image into a density perception U-Net model, extracting features through an edge guide branch and a material perception branch, generating an attention map and strengthening key features, and outputting an article density map and a bounding box set by a detector; processing the density map based on a density peak detection algorithm to obtain a predicted object number, and performing error correction through multi-angle projection reconstruction and density map fusion; and the counting result and the defect alarm are visually displayed. According to the invention, articles in the box body can be counted without opening the box and unsealing, and the problem of article counting in a complex scene is effectively solved.