Freezer car goods full-automatic loading and unloading monitoring and real-time goods quantity tracking method

Abstract

The invention relates to the technical field of refrigerator car cargo quantity monitoring, and discloses a refrigerator car cargo full-automatic loading and unloading monitoring and real-time cargo quantity tracking method. The method comprises the steps that multiple types of intelligent sensing devices are deployed, and cargo weight signals, position signals and loading and unloading action signals of a loading and unloading area of the refrigerator car are collected; extracting relevance characteristics of cargo weight and position signals, and adjusting characteristics of loading and unloading events on the two types of signals; on the basis of the features, generating state features representing cargo quantity states, and obtaining target cargo quantity probability distribution; performing interference suppression on the original sensing signal by using the probability distribution, and separating a target weight waveform and a target position signal of the target cargo; and finally, real-time goods quantity tracking of the goods is realized through a cloud big data analysis platform in combination with GPS / Beidou positioning data. According to the method, through multi-source signal fusion, feature deep mining and interference suppression processing, full-automatic monitoring and cargo quantity dynamic tracking in the cargo loading and unloading process of the refrigerator car are achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of refrigerated truck cargo volume monitoring, and in particular to a method for fully automatic cargo loading and unloading monitoring and real-time cargo volume tracking of refrigerated trucks. Background Art

[0002] In modern logistics systems, refrigerated trucks serve as the core transport vehicle for perishable goods such as fresh produce and pharmaceuticals. Their cargo loading and unloading efficiency and cargo tracking accuracy directly impact cargo quality assurance and the smooth operation of the supply chain. With the growing market demand for cold chain logistics, traditional refrigerated truck cargo loading and unloading and cargo volume management models are gradually exposing their limitations. Currently, most refrigerated truck cargo loading and unloading monitoring still relies on manual recording or single-device testing, which is not only time-consuming and labor-intensive, but also prone to data distortion due to human error. During cargo loading and unloading, existing technologies often rely on manual counting or simple weighing devices to calculate cargo volume, making it difficult to capture dynamic changes during the loading and unloading process in real time. For example, manual recording methods are susceptible to factors such as fatigue and negligence, resulting in inaccurate recording of key information such as cargo quantity and weight. A single weighing device can only capture the total weight of the cargo, making it impossible to distinguish the distribution of cargo in different areas, and even more difficult to correlate loading and unloading actions with changes in cargo volume. This limitation is particularly prominent when multiple batches and categories of goods are mixed, which can easily lead to deviations in cargo volume statistics and cause problems for subsequent warehousing scheduling and distribution planning. Refrigerated truck transportation environments are complex, with vibrations, temperature fluctuations, and electromagnetic interference often present in loading and unloading areas. This makes sensor signals susceptible to interference, further reducing the reliability of cargo volume detection. Traditional methods lack effective signal processing mechanisms and are unable to specifically suppress interference components in the original sensor signals, resulting in large errors in the extracted cargo weight and location information. Furthermore, the cargo volume tracking process lacks deep integration with the positioning system, making it difficult to achieve dynamic monitoring of cargo throughout the entire transportation process. When cargo is shifted, lost, or the volume is abnormal, it is impossible to detect and issue warnings in a timely manner, increasing cargo loss and management risks. With the development of intelligent logistics, the demand for intelligent and automated loading and unloading of refrigerated trucks and cargo tracking is increasing. Existing technologies lack the ability to integrate multi-source signal fusion, dynamic feature extraction, interference suppression, and real-time data analysis, resulting in refrigerated truck cargo management remaining at a semi-automated stage, making it difficult to meet the efficient and accurate demands of modern cold chain logistics. Therefore, developing a method that enables fully automated loading and unloading monitoring and real-time cargo tracking has become a key area of ​​technological advancement for cold chain logistics. Summary of the Invention

[0003] The object of the present invention is to provide a method for fully automatic loading and unloading monitoring and real-time cargo volume tracking of a refrigerated truck, so as to solve the problems raised in the above-mentioned background technology.

[0004] To achieve the above objectives, the present invention provides a method for fully automatic cargo loading and unloading monitoring and real-time cargo volume tracking of a refrigerated truck, the method comprising: Deploy multiple types of intelligent sensing equipment to collect raw sensor signals from the refrigerated truck loading and unloading area, and extract cargo weight signals, cargo location signals, and action signals corresponding to loading and unloading events from the raw sensor signals; Acquire a correlation characteristic between the cargo weight signal and the cargo location signal, and a regulation characteristic of the cargo weight signal and / or the cargo location signal caused by the loading and unloading event; generating a state feature representing a cargo quantity state according to the correlation feature and the adjustment feature; and obtaining a target cargo quantity probability distribution based on the state feature; performing interference suppression processing on the original sensor signal according to the target cargo quantity probability distribution to separate the target weight waveform and the target position waveform of the target cargo from the original sensor signal; Combined with GPS / Beidou positioning data, real-time cargo tracking is carried out through the cloud-based big data analysis platform.

[0005] Preferably, obtaining the correlation characteristics between the cargo weight signal and the cargo location signal, and the adjustment characteristics of the cargo weight signal and / or the cargo location signal caused by the loading and unloading event, comprises the following steps: Calculate the weight change rate and position-weight correlation coefficient during the loading and unloading cycle to obtain correlation characteristics; Establishing a loading and unloading-position coordination criterion, analyzing the stability index and weight recovery rate of the cargo position signal before and after the loading and unloading event to obtain a regulation characteristic; The correlation feature is used to adjust the generation process of the target cargo volume probability distribution, and the adjustment feature is used to optimize the interference suppression process.

[0006] Preferably, extracting the action signal corresponding to the loading and unloading event from the original sensor signal includes the following steps: Performing energy gradient analysis on adjacent signal frames of the original sensor signal to locate candidate loading and unloading intervals and generate loading and unloading event markers; Performing non-stationary feature decoupling on the marker signal of the loading and unloading event marker to separate the frequency band feature components that characterize the loading and unloading action; Distinguishing local loading and unloading actions from global interference based on motion trajectory discrimination criteria, and fusing the frequency band feature components to generate an action intensity waveform; The action intensity waveform is used to assist in generating the adjustment feature, and the loading and unloading event mark is used to trigger the calculation of the correlation feature.

[0007] Preferably, performing energy gradient analysis on adjacent signal frames of the original sensor signal, locating candidate loading and unloading intervals, and generating loading and unloading event markers comprises the following steps: Calculating a Teager energy operator difference sequence of adjacent signal frames of the original sensor signal through a sliding time window; According to the real-time signal characteristics, a dynamic judgment threshold is obtained; According to the dynamic determination threshold, the starting time point and duration range of the loading and unloading event are obtained; Locating the loading and unloading candidate interval based on the starting time point and duration range of the loading and unloading event, and generating a loading and unloading event marker; The output of the loading and unloading event mark is used to control the separation process of the frequency band characteristic components.

[0008] Preferably, the method of distinguishing local loading and unloading actions from global interference based on motion trajectory discrimination criteria and fusing the frequency band characteristic components to generate an action intensity waveform comprises the following steps: Construct a three-dimensional motion trajectory map based on Doppler phase changes and calculate the Mahalanobis distance between each trajectory point and the historical motion reference trajectory; When the cumulative Mahalanobis distance of continuous trajectory points exceeds the dynamic interference threshold, it is determined as a local loading and unloading action event and the spatial positioning coordinates are output; Performing spatiotemporal alignment of the instantaneous energy of the frequency band characteristic component with the spatial positioning coordinates to generate a time-domain continuous motion intensity waveform; The spatial positioning coordinates are used to optimize the establishment of the loading and unloading-position coordination criterion, and the data stream of the action intensity waveform is used to update the analysis of the weight recovery rate.

[0009] Preferably, after obtaining the correlation feature between the cargo weight signal and the cargo position signal, and the adjustment feature of the loading and unloading event on the cargo weight signal and / or the cargo position signal, and before generating the status feature representing the cargo quantity status based on the correlation feature and the adjustment feature, the following steps are further included: Determine the duration range of the loading and unloading event and the size of the preset threshold, and determine whether there is overlap of multiple target reflection cross sections; Whether to perform secondary processing is selected according to the judgment result. The judgment of the overlap of the multi-target reflection cross sections is based on the output of the action intensity waveform generated by distinguishing local loading and unloading actions from global interference and fusing frequency band characteristic components.

[0010] Preferably, after determining the difference between the duration range of the loading and unloading event and the preset threshold value, and determining whether there is overlap of multiple target reflection cross sections, the method further includes the following steps: If the duration of the loading and unloading event is greater than a preset threshold, or there is overlap of multiple target reflection cross sections, a multi-source data fusion algorithm is used to perform secondary decoupling processing on the cargo weight signal and the cargo position signal; Based on the three-dimensional motion trajectory map, the overlapping areas of the reflection cross sections corresponding to the global interference are excluded, and the processed weight signal and the processed position signal are re-extracted; Obtaining a processed correlation feature of the target individual based on the processed weight signal; obtaining a processed adjustment feature based on the processed position signal; and using the processed correlation feature and the processed adjustment feature to optimize the generation of the state feature; The data inputs of the processed weight signal and the processed position signal are used to update the calculation of the target cargo quantity probability distribution.

[0011] Preferably, after performing secondary decoupling processing on the cargo weight signal and the cargo position signal, the method further includes the following steps: Verify the stability of signal energy distribution after secondary decoupling processing; If the verification is successful, the state feature is modified based on the post-processing correlation feature and the post-processing adjustment feature to obtain a post-processing feature, and the post-processing feature is used as an input for generating the target cargo volume probability distribution; The verification result of the signal energy distribution stability is used to control the parameter adjustment of the interference suppression processing.

[0012] Preferably, the real-time cargo tracking through the cloud big data analysis platform includes the following steps: When the sensor data traffic exceeds the threshold and the cloud resource utilization continues to rise, the global load value is generated by calling multi-source sensor data and constructing a time series dataset; When the global load value reaches the peak threshold preset by the system, the priority and resource consumption of the analysis model are evaluated based on the scheduling coefficient, and the high-priority model is assigned to the active state; Perform resource recycling or preloading based on the value accumulation function and model switching overhead, and reserve cloud resource quotas based on the recent value accumulation function prediction results; The output of the global load value is used to dynamically adjust the calculation of the scheduling coefficient, and the data of the reserved cloud resource quota is used to support the continuity of the real-time cargo tracking.

[0013] Preferably, the method further includes the following steps before generating the global load value after calling the multi-source sensor data and constructing the time series data set: Based on the acquired original sensor signal, the target modal functions are generated through adaptive decomposition; Eliminate the modal function with the largest total entropy value from the target modal functions to obtain the effective signal; Identify abnormal data by comparing discrete data based on valid signals, remove and replace abnormal data to obtain new valid signals; Iterative processing is performed until the iteration cutoff condition is reached to generate a denoised signal; The output of the denoised signal is used to construct the time series data set, and the identification result of the abnormal data is used to update the evaluation of the value accumulation function.

[0014] Compared with the prior art, the present invention has the following beneficial effects: This method overcomes the limitations of traditional single-sensor approaches by deploying multiple types of intelligent sensors. It can simultaneously collect cargo weight, position, and loading and unloading motion signals, enabling comprehensive capture of multi-dimensional data during the cargo loading and unloading process. This simultaneous acquisition of multiple signals eliminates the reliance on isolated data to describe cargo status, providing a more complete information foundation for subsequent analysis through the rich raw signals. During feature extraction, by exploring the correlation between cargo weight and position signals, as well as the modulation of these two types of signals by loading and unloading events, we can more accurately characterize the dynamic changes in cargo during loading and unloading. This feature-correlation-based analysis avoids over-reliance on a single signal, making the representation of cargo status more consistent with actual loading and unloading scenarios and reducing judgment bias caused by signal isolation. Generating state features that characterize the cargo volume and obtaining a probability distribution for the target cargo volume provides a scientific basis for subsequent signal processing. This probability distribution model distinguishes valid information from interference components in the original sensor signal. Interference suppression processing based on this model effectively separates the weight waveform and position waveform of the target cargo, reducing the impact of environmental interference, equipment noise, and other factors on signal accuracy, ensuring that the extracted cargo parameters are more accurate than actual conditions. Combining GPS / Beidou positioning data with a cloud-based big data analysis platform enables real-time tracking and dynamic management of cargo volume information. Positioning data provides a spatial reference for cargo volume information, while the cloud-based platform enables centralized data processing and real-time updates. This allows logistics managers to monitor changes in refrigerated truck cargo volume during transportation and promptly detect any abnormal fluctuations in cargo status, providing timely information support for decisions such as cargo scheduling and route optimization. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 This is a working principle diagram of the method for fully automatic loading and unloading monitoring and real-time cargo volume tracking of refrigerated trucks according to the present invention; Figure 2 Flowchart for extracting action signals corresponding to loading and unloading events; Figure 3 Flowchart generated for locating candidate loading and unloading intervals and marking loading and unloading events; Figure 4 This is a flow chart for secondary processing and judgment of loading and unloading events. DETAILED DESCRIPTION

[0016] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0017] See also Figure 1 The present invention provides a method for fully automatic loading and unloading monitoring and real-time cargo volume tracking of refrigerated trucks, the method comprising: The implementation of fully automated refrigerated truck loading and unloading monitoring and real-time cargo tracking begins with the deployment of a variety of intelligent sensing devices within the truck's loading and unloading area. These devices include, but are not limited to, high-precision weighing sensors, such as strain gauge sensors integrated into the loading platform or floor, spatial positioning sensors, such as ultra-wideband radar (UWB) radar, lidar, or multi-camera vision systems, and sensors for capturing loading and unloading motions, such as inertial measurement units (IMUs), vibration sensors, or acoustic sensors operating in specific frequency bands. These sensors work together to continuously collect raw sensor signals from the loading and unloading area. Three key pieces of information must be extracted from these raw sensor signals: cargo weight signals, reflecting real-time changes in cargo mass; cargo position signals, reflecting the cargo's three-dimensional spatial coordinates within the truck; and motion signals corresponding to loading and unloading events, reflecting the dynamics of loading and unloading machinery or personnel. Weight signals typically come from weighing sensors, position signals from spatial positioning sensors, and motion signals extracted through specialized analysis of raw signals, such as vibration, sound waves, or specific image features.

[0018] After acquiring the aforementioned signals, the system then analyzes the inherent correlation between changes in cargo weight and its position within the carriage during loading and unloading. Furthermore, it analyzes the modulation characteristics of the loading and unloading events on the cargo weight or position signals. This refers to how the loading and unloading actions themselves affect the fluctuation pattern of the weight signal, such as how the impact of loading and unloading moments can cause temporary misalignment or drive significant changes in the position signal, such as when cargo is moved. Based on the extracted correlation and modulation characteristics, the system generates state features that characterize the cargo volume. These state features provide a comprehensive, quantitative description of the current cargo load, distribution, and the impact of loading and unloading activities. Using these state features, the system further derives a probability distribution for the target cargo volume. This probability distribution model describes the likelihood of the actual cargo volume within the carriage being in different states, given sensor observations and characteristic conditions, providing a statistical basis for subsequent processing.

[0019] The system performs interference suppression on the raw sensor signals based on the probability distribution of the target cargo volume. This process utilizes a probabilistic model to identify and filter out noise or interference components in the raw signals that are unrelated to the target cargo's state, such as slight vehicle sway, ambient vibration, and sensor noise floor. After interference suppression, the target cargo's target weight waveform and target position waveform can be separated from the raw sensor signals. These waveforms more clearly and accurately reflect the target cargo's weight changes and spatial movement trajectory. Combined with the refrigerated truck's GPS / Beidou positioning data, the system provides information on the vehicle's geographic location and motion status. All processed signals and feature data are uploaded to a cloud-based big data analysis platform. Leveraging powerful computing power and historical data resources, the cloud integrates, mines, and analyzes data from multiple refrigerated trucks and multiple loading and unloading operations. This enables real-time tracking of cargo loads, location distribution, and changes within refrigerated trucks, providing real-time decision support for logistics management, inventory optimization, and transportation scheduling.

[0020] Example 1 See Figure 2 Refrigerated trucks are equipped with an array of high-precision weighing sensors located in key load-bearing areas of the floor. These sensors collect real-time data on changes in pressure exerted by the cargo, forming the primary source of cargo weight signals. The spatial positioning sensor network deployed on the roof and side walls of the vehicle utilizes ultra-wideband radar technology, supplemented by deep vision cameras, to continuously scan and acquire the three-dimensional point cloud coordinates of the cargo's outer contour, forming the core data stream for cargo location signals. Broadband vibration sensor arrays and directional microphones positioned around the loading and unloading areas capture the physical vibrations and sound waves of specific frequencies generated by mechanical operations. These raw physical quantities constitute the source information for loading and unloading event analysis. The system uses a unified timestamp to synchronize the acquisition of all sensor data streams, ensuring traceable temporal correlation between different signals. Minutes before loading and unloading operations begin, the system automatically performs a baseline calibration, recording the zero offset and ambient noise characteristics of each sensor while the vehicle is stationary. This generates a dynamic background model for subsequent signal extraction.

[0021] When continuously acquiring raw sensor data streams, the system's primary task is to separate the mixed physical quantities into three recognizable signal types. For the continuous voltage signal collected by the pressure sensor array, the system removes low-frequency baseline drift caused by vehicle engine idling by employing an adaptive high-pass filter to eliminate fluctuations below a specific cutoff. Spikes in the remaining signal that exceed the ambient noise average typically correspond to transient pressure fluctuations caused by the placement or removal of cargo. The system identifies these fluctuations by comparing the differences in signal energy integrals within adjacent time windows. The key to spatial positioning signal processing lies in removing static interfering reflection points. A dynamic clustering algorithm is used to identify valid point clusters belonging to the cargo using multi-frame point cloud data. Position trajectory data is generated by calculating the change in the point cluster's centroid in the three-dimensional coordinate system. The raw signals from the vibration sensors undergo multi-stage bandpass filtering to retain energy variations in frequency bands reflecting metal impact and motor operation. The acoustic sensors focus on analyzing the characteristic frequency bands of cargo collision and conveyor belt friction. When multiple sensor channels detect synchronous changes consistent with loading and unloading characteristics at the same timestamp, the system generates an event marker with a confidence score and records the event trigger time. Correlation analysis between weight and position signals requires addressing the issue of physical dimension differences. The system introduces a dynamic time warping algorithm to align weight mutation points and position jump points on the time axis. The position-weight synchronization correlation coefficient is calculated by counting the frequency of successful matches within a preset time window. The modulating effect of loading and unloading events on the signal is reflected in the recovery characteristics of the weight sensor reading from shock fluctuations to stable values. The system uses the time it takes for the signal standard deviation to converge to a baseline level as an indicator of weight recovery rate. Position stability is quantified by the ratio of the change in the distribution radius of the cargo point cloud before and after the event.

[0022] The application of correlation features directly influences the construction of the target cargo volume probability model. When the position-weight synchronization coefficient remains below a threshold, the system automatically reduces the weight of the position signal in the joint probability distribution model. Otherwise, the confidence parameter of the position dimension is increased. The recovery rate parameter in the adjustment feature controls the attenuation coefficient of the probability distribution function. This broadens the distribution function during periods of slow recovery after loading and unloading, enhancing tolerance to potential residual weight interference. When generating cargo volume status features, the system employs a multidimensional feature fusion mechanism to encode the weight mean, position distribution entropy, and loading and unloading action intensity into a fixed-dimensional vector descriptor. The state feature vector is input into a cargo volume probability estimator based on a Gaussian mixture model. The probability density distribution of the current cargo state in the cargo volume space is generated based on the cluster centers of historical training data. This distribution guides the filtering of the raw signal using a variable bandwidth suppression strategy: narrowband filtering preserves details in high-probability density regions and wideband filtering eliminates suspicious fluctuations in low-probability regions. The key to isolating the target weight waveform lies in constructing adaptive filter coefficients weighted by the probability distribution to effectively suppress atypical weight variation patterns. Optimizing the position waveform relies on evaluating the plausibility of the trajectory using a probability distribution. During the position coordinate update phase, constraints on cargo volume changes are introduced to eliminate physically inconsistent drift points. After all signals are processed, the onboard gateway generates a time-stamped feature data packet containing the target weight waveform sampling points, the target position coordinate sequence, and key feature statistics.

[0023] Energy gradient analysis of raw sensor signals requires a balance between detection sensitivity and noise immunity. The system utilizes a dual dynamic threshold mechanism to process energy changes within a sliding window. The first threshold, set as a multiple of the RMS value of the ambient noise, is used to initially screen suspected event segments. The second threshold is dynamically adjusted based on the smoothness of the multi-second signal, automatically raising the threshold for event triggering on bumpy roads. The time window size must encompass baseline data for the complete loading and unloading cycle. The initial window value is set based on the typical forklift operating rhythm and automatically scales up and down based on operating conditions. When the energy change in a specific channel exceeds the dual thresholds, the system records the starting timestamp and continuously monitors until the signal fluctuations return to a stable range, denoting the end of the signal fluctuation. When the event duration data is written to the dynamic database, an abnormal status flag is set to facilitate subsequent processes in determining whether to trigger secondary processing mechanisms. This flag also drives configuration parameter modifications for the preprocessing process of the next stage of signal decomposition, determining the spatial sampling accuracy of the non-stationary feature decoupling algorithm. The core of non-stationary feature decoupling lies in separating the intrinsic signal components representing loading and unloading movements.

[0024] The system uses variational mode decomposition to decompose the multi-source sensor data stream during the marking period into a finite number of narrowband oscillatory components. The clustering results of the center frequencies of each component correspond to typical acoustic characteristics of the operation. For example, components concentrated between 250 and 450 Hz are identified as resonance between the fork tines and the shelf; high-frequency components are used to identify the tensile properties of the plastic wrap. The motion trajectory identification process reconstructs the cargo displacement vector using the raw phase change matrix of the millimeter-wave radar and detects abnormal motion patterns by comparing it with a pre-set coordinate system reference grid. The dynamic trajectory point set of spatial position information is evaluated for compatibility with a pre-set standard loading and unloading path library, and the angular deviation and velocity fluctuation index of each path node are calculated. Consecutive abnormal trajectory points are clustered using spatial density to identify local loading and unloading event hotspots. The output of the hotspot location coordinates simultaneously corrects the regional weighting coefficient in the position stability assessment algorithm. Action intensity waveform generation quantifies the mapping of physical signals to the intensity of the operation behavior. The instantaneous energy integral values ​​of the components in each frequency band are weighted and fused into a unified intensity index using the sensor's spatial position coefficient. Spatial coordinate calibration ensures accurate positioning and mapping of the moving target within the three-dimensional scene. This waveform data periodically updates the weight recovery rate model parameters in the state feature analysis module. The system uses regression analysis to establish a relationship matrix between peak motion intensity and weight stabilization time, enabling real-time correction and adjustment of the feature. The output data also drives the cloud-based operation mode analysis engine to automatically establish a library of optimal operating benchmarks. This long-term accumulation of waveform feature data provides a critical basis for decision-making in optimizing loading and unloading procedures.

[0025] Example 2 See Figure 3Millimeter-wave radar arrays deployed in the refrigerated truck loading and unloading areas form a three-dimensional spatial perception network. Each radar node continuously transmits frequency-modulated continuous waves and receives reflected signals. The original echo signal contains static environmental reflections and dynamic target information. The system uses range-Doppler processing to separate the frequency domain feature points corresponding to the moving target. For each potential moving point detected, the system records its range frequency offset and azimuth phase difference, and calculates the target's instantaneous three-dimensional coordinates in the radar coordinate system based on the multi-antenna geometry. Coordinate points at adjacent moments are connected in chronological order to form preliminary motion trajectory segments. The system automatically removes jump points caused by multipath effects and smoothes the trajectory data using Kalman filtering. Vibration signals generated during loading and unloading operations are collected by accelerometers placed in the vehicle body structure. The original vibration waveform is bandpass filtered to retain the frequency band reflecting mechanical impact. A sliding time window is used to calculate the Teager energy operator value of the signal within each window. The teager energy difference sequence between adjacent time windows is calculated in real time. When the difference exceeds a threshold dynamically adjusted based on the ambient vibration level, an event start marker is triggered. The system continuously monitors until the difference falls below the threshold to determine the event duration and generate a loading and unloading event marker segment containing start and end timestamps. This marker segment controls resource allocation for subsequent signal processing, activating the computationally complex feature extraction algorithm only during the marker period.

[0026] During the marking period, the system reconstructs the three-dimensional motion trajectory of the millimeter-wave radar raw data. By solving the phase difference matrix △Φ between each receiving antenna, combined with the carrier wavelength λ and the antenna spacing d, the target azimuth θ is calculated: ; Where: θ represents the target azimuth, λ is the carrier wavelength of the radar signal, △Φ is the phase difference matrix between the receiving antennas, and d is the antenna spacing. The integrated multi-target tracking algorithm associates the spatial points of consecutive frames to form a time-varying trajectory point set P(t)=[x(t), y(t), z(t)]. In order to distinguish between local loading and unloading actions and the overall movement of the vehicle, the system establishes a historical trajectory benchmark model: during the period without loading and unloading operations, trajectory data under typical driving conditions is collected, and the first three principal component directions are extracted through principal component analysis to construct a benchmark subspace B. For the new trajectory point P(t), calculate its Mahalanobis distance D to the benchmark subspace B. M (P): ; in: is the mean vector of the reference trajectory, is the covariance matrix of the reference trajectory, and the new trajectory point P. When the cumulative Mahalanobis distance of five consecutive trajectory points exceeds the dynamic threshold, it is determined to be a valid loading and unloading action, and the spatial positioning coordinates of the action are output C=[x c ,y c ,z cThe vibration sensor signals are processed synchronously, and variational modal decomposition is performed on the acceleration data within the marked period to isolate the specific frequency band components that represent the loading and unloading action. The instantaneous energy of each component is spatiotemporally aligned with the spatial positioning coordinate C to generate a time-domain continuous action intensity waveform. The dynamic adjustment mechanism of the key parameters involved in the action intensity waveform generation process is based on real-time working condition feedback.

[0027] The spatial positioning coordinates C update the regional weight coefficients used in the loading and unloading-position coordination criterion in real time. When loading and unloading activity is detected to be concentrated in the front area of ​​the vehicle, the position stability assessment algorithm automatically increases the data sampling frequency of the front sensor. The motion intensity waveform data stream is fed into the weight recovery analysis module, where the system develops a regression model between intensity peaks and weight signal stabilization time. When the monitored intensity peak exceeds two standard deviations of the historical mean, the calculation window for the weight recovery rate is automatically extended. The accuracy of the three-dimensional motion trajectory map is improved through multi-sensor fusion. UWB anchor points added at the four corners of the vehicle provide auxiliary positioning information, optimizing trajectory coordinate accuracy by minimizing reprojection error. The benchmark model required for Mahalanobis distance calculation is automatically updated every ten minutes to adapt to changes in vehicle vibration patterns caused by changing road conditions. The spatiotemporal alignment phase uses dynamic time warping to address time delays between vibration signals and spatial coordinates, ensuring that the motion intensity waveform accurately reflects the spatiotemporal characteristics of the physical event. The output data also drives a cloud-based analysis platform to construct a digital twin model of the loading and unloading operation. The mapping between the motion intensity waveform and spatial coordinates provides a quantitative basis for optimizing loading and unloading path planning.

[0028] Example 3 See Figure 4 After completing the preliminary signal feature extraction, the refrigerated truck loading and unloading monitoring system enters the complex working condition judgment stage. The system presets a reference value for the standard duration threshold of loading and unloading events. This value comes from the average duration of conventional single cargo handling operations recorded in the historical operation statistics database. When the duration of a loading and unloading event detected in real time exceeds 1.5 times the reference value, the system automatically marks it as an abnormally long event. The motion intensity waveform analysis module synchronously outputs a spatial heat map. When a specific area continuously experiences high-intensity pulses during the event period and the spatial coordinate fluctuation range is less than 0.3 meters, it is determined that there is a phenomenon of multi-target reflection cross-section overlap. The system confirms the judgment result through a dual-channel verification mechanism: the weight signal channel detects a continuous step-like change pattern, and the position signal channel shows a multi-point cloud cluster aggregation feature. The judgment result triggers the update of the secondary processing flag, which directly controls the branch selector of the signal processing pipeline.

[0029] The multi-source data fusion algorithm uses a hybrid architecture of blind source separation and beamforming to process abnormal event signals. The raw data stream of the weight sensor array and the millimeter wave radar point cloud data stream are input into the joint processing unit. The signal mixing model is established: X(t)=[W(t);P(t)]=A·S(t)+N(t); where X(t) represents the measurement matrix consisting of the weight signal W(t) and the position signal P(t), A is the unknown mixing matrix, S(t) is the source signal matrix to be separated, and N(t) is the ambient noise term. Independent component analysis is used to iteratively estimate the separation matrix, ensuring that the components of the output signal are statistically independent. The three-dimensional motion trajectory map provides spatial constraints. The system calculates the spatiotemporal correlations between each separated component and the trajectory hotspots, selecting components with correlations exceeding 0.7 for signal reconstruction while suppressing components corresponding to global interference regions in the trajectory map. Adaptive beamforming technology is incorporated into the reconstruction process to enhance the target direction signal gain in the spatial domain. This process yields separated target weight and position signals, with data quality quantitatively assessed by the improvement in signal-to-noise ratio.

[0030] After secondary decoupling, the signal enters the feature recalculation process: the position-weight correlation coefficient is calculated using the ratio of covariance to standard deviation, where the weight signal uses the decoupled data and the position signal uses the vertical component. When re-extracting the loading and unloading adjustment features, the action intensity waveform guidance mechanism is introduced: taking the peak moment of the action intensity waveform as the reference point, the position signal variance change rate and weight signal convergence time in the 0.5-second window before and after are analyzed. The processed correlation features and adjustment features are input into the feature optimizer, which performs feature weighted fusion based on weight coefficients learned from historical data to generate an enhanced state feature vector. The target cargo volume probability distribution model receives this feature vector as a conditional input and updates the probability density function parameters. The processed weight signal and position signal are synchronously input into the data pool, triggering the online learning mechanism of the probability distribution model to adjust the distribution morphology parameters.

[0031] Signal energy distribution stability verification utilizes a multi-index cross-validation scheme. Time-domain analysis calculates the fluctuation of the processed weight signal during the loading and unloading interval, while frequency-domain analysis extracts the energy proportion in specific low-frequency bands. A comprehensive stability index is established, calculated as the weighted sum of the inverse of the time-domain fluctuation and the low-frequency energy proportion. Verification is considered successful when the index exceeds an empirical threshold. Upon passing verification, the system marks the enhanced state features as valid input, replacing the original state features for cargo volume probability distribution calculation. The verification results are also fed into the interference suppression parameter controller, which selects different filtering strategies based on the index value range. Basic filtering parameters are selected when the index is in the high range; enhanced filtering is enabled when it is in the mid-range range; and when it falls below the critical value, three-stage filtering is triggered and the sensor diagnostics process is initiated. Parameter adjustments are implemented by gradually reducing the process noise covariance matrix and increasing the sliding average window ratio. The entire processing flow forms a closed-loop quality control mechanism, ensuring signal processing reliability under complex operating conditions.

[0032] Example 4 After performing secondary decoupling, the refrigerated truck cargo loading and unloading monitoring system enters the quality verification phase. The decoupled weight signal data stream and position signal data stream are input into the verification module, where the system performs time-domain statistical analysis on the weight signal. During the stable period between two consecutive loading and unloading events, a sequence of continuous sampling points of at least five seconds is selected. The dispersion index of the weight signal sample values ​​during this period is calculated. This index reflects the natural fluctuation level of the signal in the absence of operational interference. Frequency-domain energy distribution analysis is also performed. A fast Fourier transform is applied to the weight signal to obtain a spectrum. The ratio of vibration energy in the frequency range of 1 Hz to 5 Hz to the total energy in the entire frequency band is calculated. A high proportion of low-frequency energy typically indicates the presence of unfiltered residual mechanical vibration. Position signal verification focuses on the convergence of spatial coordinates. Within a three-second window after the loading and unloading operation, the contraction rate of the cargo point cloud distribution radius is calculated. This rate is obtained by linear regression fitting the slope of the distribution radius over time. The system sets the verification sampling frequency to half the original sensor sampling rate to balance computational load and feature fidelity requirements.

[0033] Signal energy distribution stability verification utilizes a multi-dimensional indicator fusion strategy. The system constructs a verification matrix encompassing time-domain fluctuation, frequency-domain energy, and spatial convergence metrics. The time-domain fluctuation metric is normalized using the inverse of the weight signal's variance during stable periods. The frequency-domain energy metric directly uses the raw energy percentage of the 1-5 Hz frequency band. The spatial convergence metric takes the absolute value of the rate of change of the location point cloud distribution radius. These three metrics are multiplied by adaptive weighting coefficients and summed to generate a comprehensive stability index. The weighting coefficients are dynamically adjusted based on the current cabin temperature and vehicle motion. Verification thresholds are set based on a database of historical success cases, and the system automatically updates the threshold baseline every 24 hours. When the comprehensive stability index exceeds the dynamic threshold, the secondary decoupling process is deemed successful. The verification result triggers a state feature update command, marking the enhanced state feature vector as a valid data source and replacing the state feature generated by the original processing channel. When the new feature vector is input into the target cargo volume probability distribution model, the system simultaneously records the feature switching timestamp and version identifier, providing a data anchor for subsequent quality traceability.

[0034] The interference suppression parameter adjustment mechanism establishes a three-level response mode. The system maintains a parameter configuration table that stores filter parameter sets corresponding to different verification results. This table is periodically expanded as the vehicle's cumulative operating time increases. When the comprehensive stability index is within the high confidence interval, the system selects the basic filter parameter set: the Kalman filter process noise covariance matrix uses standard configuration values, and the sliding average window length is set to the default ten sampling periods. When the medium confidence interval is triggered, the system activates enhanced filtering mode: each element of the process noise covariance matrix is ​​uniformly reduced by 30%, and the sliding average window is extended to 15 sampling periods. When the index falls below the critical threshold, the system activates a three-level response: the process noise covariance matrix is ​​halved, the sliding average window is extended to 25 sampling periods, and a sensor calibration request is simultaneously sent to the on-board diagnostic system. The parameter switching process uses a gradual adjustment algorithm to avoid signal jumps, and the new parameters are gradually implemented through linear interpolation within five sampling periods. All parameter adjustment events are logged, including fields such as timestamp, previous parameter value, target parameter value, and adjustment duration. See Table 1.

[0035] Table 1: Secondary decoupling signal verification indicator parameter table.

[0036]

[0037] The verification process includes an exception handling branch. If the comprehensive stability index fails verification three times in a row, the system automatically initiates a signal tracing program. This program retrieves cached raw sensor signal data from the five most recent loading and unloading events, re-executes the decoupling process, and records intermediate variables. The diagnostic engine compares the differences in characteristic parameters from each processing step, focusing on key node data such as the mixing matrix estimate, the number of independent components criterion, and the beamforming steering vector. Once generated, the diagnostic report is automatically uploaded to the cloud-based analysis platform, and the local system reverts to basic processing mode without secondary decoupling enabled. In fallback mode, the system attempts to reactivate the decoupling processing module every two hours until the parameter correction plan issued by the cloud is successfully loaded. The entire verification mechanism and parameter adjustment system form a closed-loop control loop, dynamically optimizing the signal processing chain configuration by continuously monitoring processing quality. All operation logs and diagnostic reports are stored chronologically in the on-board black box memory, supporting offline analysis of up to 30 days of operating data.

[0038] Example 5 The cloud platform of the refrigerated truck real-time cargo tracking system continuously monitors data access and computing resource status. When sensor data traffic exceeds preset warning levels and CPU utilization shows a continuous upward trend, the platform automatically activates a load balancing program. This program draws on multi-source sensor data streams from a distributed storage cluster, including target weight waveform sampling sequences, spatial position coordinate time series, vehicle positioning information, and motion intensity waveform data. This heterogeneous data is aligned and concatenated using millisecond-level timestamps to construct a multivariate time series dataset with a unified timeline. Based on the current number of parallel processing tasks, the data throughput of each task, and the computational complexity model, the platform uses a weighted summation algorithm to generate a global load value, which reflects the overall computing pressure level of the cloud cluster in real time. This global load value is updated every five seconds and written to the resource monitoring log, triggering subsequent scheduling decision-making processes.

[0039] When the global load reaches a preset peak threshold, the cloud-based scheduling engine initiates a dynamic resource allocation mechanism. The scheduling engine maintains a task queue containing all active analytical models and calculates a dynamic scheduling coefficient for each model instance. This coefficient is a weighted composite of four factors: a task urgency factor dynamically assigned based on the perishability level and alarm status of the goods; a data timeliness factor calculated based on a decay function of the data generation timestamp; task type weights reference a preset priority table; and historical resource consumption patterns are predicted using a machine learning model. The engine regularly evaluates the scheduling coefficient and resource consumption estimates for each model instance, sorting actively running model instances in descending order of scheduling coefficient. When the global load reaches a critical level, the system automatically migrates several high-priority model instances to a cluster of high-performance computing nodes, while simultaneously placing the lowest-priority model instances in a dormant state or migrating them to cold storage. Model state transitions are logged, including the migration timestamp and the source and target node identifiers.

[0040] The cloud platform employs a value-driven resource recycling and preloading strategy to maintain service continuity. The system constructs a value accumulation function for each model instance, dynamically calculating the expected return of the instance's continued operation over time. Parameters used in this calculation include scheduling coefficients, data freshness, and user service-level agreement weights. The resource recycling module periodically scans for model instances with low value accumulation and, based on the time cost difference between cold and hot start times, decides whether to immediately release resources or suspend them. The preloading module, based on recent trends in the value accumulation function, predicts high-value tasks likely to arrive within the next two minutes and reserves memory and processor threads on compute nodes in advance. Reserved resource quotas are set based on prediction confidence levels: high-confidence predictions are allocated a fixed-quota resource pool, while medium-confidence predictions utilize flexible resource quotas. Global load values ​​are fed back to the scheduling coefficient calculation engine in real time, automatically increasing the weight of the task urgency factor when load remains consistently high. The reserved resource data stream is connected to the load monitoring system to ensure that high-priority tasks immediately receive computing resources upon arrival, avoiding tracking delays caused by task queue congestion. All resource scheduling events generate time-stamped operation records, supporting end-to-end performance auditing and anomaly diagnosis.

[0041] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0042] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A method for fully automatic loading and unloading monitoring and real-time cargo volume tracking of refrigerated trucks, characterized in that: The following steps are involved: Deploy multiple types of intelligent sensing equipment to collect raw sensor signals from the refrigerated truck loading and unloading area, and extract cargo weight signals, cargo location signals, and action signals corresponding to loading and unloading events from the raw sensor signals; Acquire a correlation characteristic between the cargo weight signal and the cargo location signal, and a regulation characteristic of the cargo weight signal and / or the cargo location signal caused by the loading and unloading event; generating a state feature representing a cargo quantity state according to the correlation feature and the adjustment feature; and obtaining a target cargo quantity probability distribution based on the state feature; performing interference suppression processing on the original sensor signal according to the target cargo quantity probability distribution to separate the target weight waveform and the target position waveform of the target cargo from the original sensor signal; Combined with GPS / Beidou positioning data, real-time cargo tracking is carried out through the cloud-based big data analysis platform.

2. The method for fully automatic loading and unloading monitoring and real-time cargo volume tracking of refrigerated trucks according to claim 1 is characterized in that: Acquiring the correlation characteristics between the cargo weight signal and the cargo location signal, and the adjustment characteristics of the cargo weight signal and / or the cargo location signal caused by the loading and unloading event, comprises the following steps: Calculate the weight change rate and position-weight correlation coefficient during the loading and unloading cycle to obtain correlation characteristics; Establishing a loading and unloading-position coordination criterion, analyzing the stability index and weight recovery rate of the cargo position signal before and after the loading and unloading event to obtain a regulation characteristic; The correlation feature is used to adjust the generation process of the target cargo volume probability distribution, and the adjustment feature is used to optimize the interference suppression process.

3. The method for fully automatic loading and unloading monitoring and real-time cargo volume tracking of refrigerated trucks according to claim 2 is characterized in that: Extracting an action signal corresponding to a loading and unloading event from the original sensor signal includes the following steps: Performing energy gradient analysis on adjacent signal frames of the original sensor signal to locate candidate loading and unloading intervals and generate loading and unloading event markers; Performing non-stationary feature decoupling on the marker signal of the loading and unloading event marker to separate the frequency band feature components that characterize the loading and unloading action; Distinguishing local loading and unloading actions from global interference based on motion trajectory discrimination criteria, and fusing the frequency band feature components to generate an action intensity waveform; The action intensity waveform is used to assist in generating the adjustment feature, and the loading and unloading event mark is used to trigger the calculation of the correlation feature.

4. The method for fully automatic loading and unloading monitoring and real-time cargo volume tracking of refrigerated trucks according to claim 3 is characterized in that: The step of performing energy gradient analysis on adjacent signal frames of the original sensor signal, locating candidate loading and unloading intervals, and generating loading and unloading event markers includes the following steps: Calculating a Teager energy operator difference sequence of adjacent signal frames of the original sensor signal through a sliding time window; According to the real-time signal characteristics, a dynamic judgment threshold is obtained; According to the dynamic determination threshold, the starting time point and duration range of the loading and unloading event are obtained; Locating the loading and unloading candidate interval based on the starting time point and duration range of the loading and unloading event, and generating a loading and unloading event marker; The output of the loading and unloading event mark is used to control the separation process of the frequency band characteristic components.

5. The method for fully automatic loading and unloading monitoring and real-time cargo volume tracking of refrigerated trucks according to claim 3 is characterized in that: The method of distinguishing local loading and unloading actions from global interference based on the motion trajectory discrimination criterion and fusing the frequency band characteristic components to generate an action intensity waveform includes the following steps: Construct a three-dimensional motion trajectory map based on Doppler phase changes and calculate the Mahalanobis distance between each trajectory point and the historical motion reference trajectory; When the cumulative Mahalanobis distance of continuous trajectory points exceeds the dynamic interference threshold, it is determined as a local loading and unloading action event and the spatial positioning coordinates are output; Performing spatiotemporal alignment of the instantaneous energy of the frequency band characteristic component with the spatial positioning coordinates to generate a time-domain continuous motion intensity waveform; The spatial positioning coordinates are used to optimize the establishment of the loading and unloading-position coordination criterion, and the data stream of the action intensity waveform is used to update the analysis of the weight recovery rate.

6. The method for fully automatic loading and unloading monitoring and real-time cargo tracking of refrigerated trucks according to claim 1 is characterized in that: After obtaining the correlation feature between the cargo weight signal and the cargo position signal, and the adjustment feature of the loading and unloading event on the cargo weight signal and / or the cargo position signal, and before generating the status feature representing the cargo quantity status based on the correlation feature and the adjustment feature, the following steps are further included: Determine the duration range of the loading and unloading event and the size of the preset threshold, and determine whether there is overlap of multiple target reflection cross sections; Whether to perform secondary processing is selected according to the judgment result. The judgment of the overlap of the multi-target reflection cross sections is based on the output of the action intensity waveform generated by distinguishing local loading and unloading actions from global interference and fusing frequency band characteristic components.

7. The method for fully automatic loading and unloading monitoring and real-time cargo volume tracking of refrigerated trucks according to claim 6 is characterized in that: After determining the difference between the duration range of the loading and unloading event and the preset threshold, and determining whether there is overlap of multiple target reflection cross sections, the following steps are also included: If the duration of the loading and unloading event is greater than a preset threshold, or there is overlap of multiple target reflection cross sections, a multi-source data fusion algorithm is used to perform secondary decoupling processing on the cargo weight signal and the cargo position signal; Based on the three-dimensional motion trajectory map, the overlapping areas of the reflection cross sections corresponding to the global interference are excluded, and the processed weight signal and the processed position signal are re-extracted; Obtaining a processed correlation feature of the target individual based on the processed weight signal; obtaining a processed adjustment feature based on the processed position signal; and using the processed correlation feature and the processed adjustment feature to optimize the generation of the state feature; The data inputs of the processed weight signal and the processed position signal are used to update the calculation of the target cargo quantity probability distribution.

8. The method for fully automatic loading and unloading monitoring and real-time cargo volume tracking of refrigerated trucks according to claim 7 is characterized in that: After the secondary decoupling process is performed on the cargo weight signal and the cargo position signal, the following steps are further included: Verify the stability of signal energy distribution after secondary decoupling processing; If the verification is successful, the state feature is modified based on the post-processing correlation feature and the post-processing adjustment feature to obtain a post-processing feature, and the post-processing feature is used as an input for generating the target cargo volume probability distribution; The verification result of the signal energy distribution stability is used to control the parameter adjustment of the interference suppression processing.

9. The method for fully automatic loading and unloading monitoring and real-time cargo volume tracking of refrigerated trucks according to claim 1 is characterized in that: The real-time cargo tracking through the cloud-based big data analysis platform includes the following steps: When the sensor data traffic exceeds the threshold and the cloud resource utilization continues to rise, the global load value is generated by calling multi-source sensor data and constructing a time series dataset; When the global load value reaches the peak threshold preset by the system, the priority and resource consumption of the analysis model are evaluated based on the scheduling coefficient, and the high-priority model is assigned to the active state; Perform resource recycling or preloading based on the value accumulation function and model switching overhead, and reserve cloud resource quotas based on the recent value accumulation function prediction results; The output of the global load value is used to dynamically adjust the calculation of the scheduling coefficient, and the data of the reserved cloud resource quota is used to support the continuity of the real-time cargo tracking.

10. The method for fully automatic loading and unloading monitoring and real-time cargo volume tracking of refrigerated trucks according to claim 9 is characterized in that: After calling multi-source sensor data and constructing a time series data set and before generating a global load value, the method further includes the following steps: Based on the acquired original sensor signal, the target modal functions are generated through adaptive decomposition; Eliminate the modal function with the largest total entropy value from the target modal functions to obtain the effective signal; Identify abnormal data by comparing discrete data based on valid signals, remove and replace abnormal data to obtain new valid signals; Iterative processing is performed until the iteration cutoff condition is reached to generate a denoised signal; The output of the denoised signal is used to construct the time series data set, and the identification result of the abnormal data is used to update the evaluation of the value accumulation function.

Images