Intelligent monitoring and early warning warehouse for tea storage

Abstract

The invention discloses an intelligent monitoring and early warning warehouse for tea storage, and relates to the technical field of agricultural product storage monitoring and intelligent early warning. An acoustic respiration sensing array, a quantum dot gas mapping film and optical excitation imaging, long-wave infrared polarization insect trace detection, and an edge calculation and actuator cluster are integrated in the warehouse body; a limited scanning sector is triggered by acoustics first sense, a VOC / ammonia concentration heat map and a polarization disturbance heat map are obtained, graded early warning is output through time-space fusion consistency scoring, and low-energy-consumption zoning intervention is achieved through model prediction control cooperation dehumidification, air supply, cold plasma and a zoning air door. In cooperation with reference self-correction, channel health detection and fault-tolerant synchronization, a monitoring-decision-regulation-re-monitoring closed loop is formed, and mildew / insect pest early warning and traceable management are achieved under the condition that packaging is not damaged.

Description

Technical Field

[0001] This invention relates to the field of agricultural product storage monitoring and intelligent early warning technology, specifically to a system and warehouse structure for multimodal monitoring, graded early warning and zoned collaborative control of tea storage environment, involving the integrated application of acoustic signal acquisition and array processing, optical gas visualization imaging, long-wave infrared polarization imaging, edge computing and model predictive control, specifically a smart monitoring and early warning warehouse for tea storage. Background Technology

[0002] Even after drying, tea leaves still have a significant risk of hygroscopicity and loss of volatile components. Uneven distribution of temperature, relative humidity, and local airflow within the storage area can lead to a chain reaction of micro-area moisture accumulation—localized warming—early mold / insect activity. Inappropriate stacking methods, shelf obstructions, cold bridge locations, and ventilation can create high-risk sectors that are difficult to cover by routine monitoring. Early anomalies often occur covertly while the packaging is unopened. Once they develop to a sensory-detectable stage, batch-wide quality degradation and traceability issues are usually already difficult to address.

[0003] Current technologies generally achieve warehouse monitoring and control through temperature and humidity recorders, discrete electrochemical gas sensors, visible light camera inspections, thermal imager spot checks, and threshold-based on / off operation of dehumidifiers and fans. While these technologies can provide macro-level trend assessments of the overall warehouse environment, they still have limitations in meeting the quality control requirements for long-term tea storage. Firstly, point-based temperature, humidity, and gas sensors struggle to reflect spatial gradients within stacks and at aisle corners, making it difficult to promptly identify localized moisture accumulation and early abnormal VOC releases. Furthermore, long-term sensor drift and calibration maintenance costs are high. Secondly, visible light or single thermal imaging methods are sensitive to light, dust, and surface reflection, making it difficult to reliably confirm insect micro-movements and latent activity in heavily obstructed shelving environments. Thirdly, routine inspections rely on full-area equal-frequency scanning, resulting in significant data redundancy and energy consumption, making it difficult to achieve high-frequency area array coverage in large-capacity warehouses. Fourth, the control layer often employs single-loop or fixed-threshold open-loop logic, lacking coordinated optimization of timing and intensity between dehumidification, air supply, and disinfection, easily leading to excessive drying, condensation, and energy consumption peaks. Fifth, the monitoring system is prone to data acquisition link interruptions and gaps under conditions of communication fluctuations or power supply anomalies, making it difficult to guarantee continuous early warning and closed-loop response. Sixth, the correlation between anomaly detection and handling records and batches / locations is insufficient, resulting in incomplete evidence chains and affecting subsequent traceability and auditing.

[0004] Therefore, there is an urgent need for an intelligent monitoring and early warning warehouse for tea storage conditions. This warehouse should be able to narrow the scanning range through acoustic pre-sensing without damaging the packaging, combine quantum dot gas visualization imaging to obtain the spatial distribution of VOCs and ammonia, and use long-wave infrared polarization to identify insect bodies and activity. It should achieve highly reliable graded early warning through spatiotemporal fusion and consistency scoring, and use model predictive control to coordinate the scheduling of dehumidification, air supply, cold plasma and zoned dampers. It should also have the ability to reference self-calibration, channel health detection, edge fault tolerance and batch location correlation traceability of the warehouse management system. This will reduce energy consumption while improving the timeliness and accuracy of early warning, forming a stable closed loop of monitoring-decision-control-re-monitoring. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the prior art and propose an intelligent monitoring and early warning warehouse for tea storage to solve the above-mentioned problems.

[0006] The objective of this invention is achieved through the following technical solution: an intelligent monitoring and early warning warehouse for tea storage, comprising a warehouse body, a shelving system, a monitoring and execution subsystem, and a main control center. The monitoring and execution subsystem includes an acoustic respiration sensing array, a quantum dot gas mapping membrane, an optical excitation imaging subsystem, an infrared polarization interferometry insect trace detection component, an actuator cluster, and an edge computing unit. The acoustic respiration sensing array is installed in the aisle and corner areas of the storage unit. It includes an array of MEMS microphones and piezoelectric excitation receivers to collect low-frequency structural vibrations and gas emission acoustic responses generated by the tea pile under conditions of moisture absorption, moisture release, and internal micro-heating, and outputs abnormal respiration location information. The quantum dot gas mapping membrane is applied in a patch-like manner to the surface of the shelves and the wall area near cold bridges. The quantum dot gas mapping membrane includes, from bottom to top, a PET substrate layer, a sensitive layer formed by InP / ZnS quantum dots and molecularly imprinted polymers, and a... The system includes a light-sensitive layer and a hydrophobic microporous protective layer. An optical reference area is embedded within the quantum dot gas mapping film and managed using patch numbering. The optical excitation imaging subsystem comprises a narrowband excitation LED array, an industrial camera, a multi-channel filter assembly, and a track-mounted gimbal mechanism. This subsystem is used to perform zoned polling scanning of the area covered by the quantum dot gas mapping film and acquire optical images. The infrared polarization interference insect trace detection assembly includes a long-wave infrared polarization camera and an infrared window filter. This assembly is used to acquire polarization parameters and output polarization perturbation thermal maps. The actuator cluster includes a frequency conversion dehumidification unit, a zoned air supply unit, a cold plasma sterilization unit, and a zoned damper unit. The edge computing unit is communicatively connected to the main control center. The edge computing unit is used to preprocess the acoustic, optical, and polarization raw data and execute zoned trigger scheduling. The main control center is configured to execute the following steps: S1: Based on the array beamforming and time-frequency analysis of the acoustic breathing sensing array, generate a breathing feature map and output abnormal breathing location information; S2: Based on the abnormal breathing location information, control the optical excitation imaging subsystem to perform partitioned illumination and imaging acquisition of the corresponding covered area, and limit the scanning range to the sector marked by the abnormal breathing location information; S3: Optical comparison is performed between the reference region and the sensitive light-emitting region of the quantum dot gas mapping film to generate a heat map of the gas concentration distribution of volatile organic compounds and ammonia; S4: During the partitioned polling scan, the infrared polarization interferometric insect trace detection component is synchronously triggered to image the corresponding area, calculate the polarization parameters and generate a polarization perturbation heat map. S5: Spatial registration and consistency scoring are performed on the respiratory feature map, gas concentration distribution heat map and polarization perturbation heat map. Based on the judgment result that the consistency score exceeds the threshold, graded early warning information is output. S6: Based on the hierarchical early warning information, it sends zone control instructions to the actuator cluster to complete the coordinated scheduling of dehumidification setpoint, air supply volume and flow direction, cold plasma duty cycle and zone damper opening. S7: Based on the execution results and monitoring data, update the scanning frequency, excitation power, polarization sampling cycle and acoustic excitation amplitude, and perform reference self-calibration on the optical attenuation of the quantum dot gas mapping film, forming a closed-loop operation of monitoring, decision-making, regulation and re-monitoring.

[0007] The acoustic breathing sensing array is deployed at equal intervals along the direction of the shelf aisle. Reference nodes are set at the corners of the warehouse and the ventilation end for background noise modeling. In step S1, the main control center performs door opening and closing event elimination, transient suppression and beam pointing constraint, and outputs abnormal breathing location information including spatial positioning coordinates and abnormal intensity level.

[0008] The sensitive light-emitting layer of the quantum dot gas mapping membrane is a composite membrane formed by cross-linking ligand-stabilized InP / ZnS quantum dots with molecularly imprinted polymers for ethanol and ammonia. A hydrophobic microporous protective layer covers the composite membrane and retains gas diffusion channels. The main control center records the patch number and lifetime count and triggers a replacement prompt when the reference ratio is lower than the threshold.

[0009] The narrowband excitation LED array of the optical excitation imaging subsystem is driven by constant current pulse width modulation and operates by a zoned lighting strategy. The industrial camera is equipped with a multi-channel filter component. In step S3, the main control center performs temperature and humidity compensation based on the ratio of the intensity of the reference area to the intensity of the sensitive emitting area, and outputs a calibrated gas concentration distribution thermal map.

[0010] The infrared polarization interferometric insect trace detection component acquires the original polarization image through a long-wave infrared polarization camera. In step S4, the main control center performs multi-frame fusion, polarization parameter reconstruction and structural gradient extraction, and spatially registers it with the gas concentration distribution heat map obtained in step S3 to output a polarization perturbation heat map.

[0011] In step S5, the main control center uses a spatiotemporal graph structure data fusion model to determine the consistency of the respiratory feature map, gas concentration distribution heatmap and polarization perturbation heatmap. It also introduces virtual abnormal samples based on the generative model and real running samples for adversarial training, dynamically updates the consistency score threshold and multimodal feature weights, and outputs interpretable graded early warning information containing the trigger source and evidence map.

[0012] The actuator cluster is coordinated and scheduled by the main control center according to the model predictive control strategy. In step S6, with quality risk and energy consumption as the target constraints, the linkage control quantities of dehumidification set point, air supply volume and flow direction, cold plasma duty cycle and zone damper opening are generated. In the state of severe warning, the zone isolation and spot check task order are executed.

[0013] In step S7, the main control center performs optical self-test, acoustic self-test, and polarization self-test. The optical self-test includes sequential sampling of the quantum dot gas mapping film reference area and the fixed white board and compensation for light attenuation of the light source and camera. The acoustic self-test includes low-amplitude frequency sweep detection of the health status of the array channel. The polarization self-test includes polarization baseline correction. When an abnormality is detected, a fault alarm is triggered.

[0014] The edge computing unit is used to complete acoustic signal preprocessing, preliminary optical image calculation and preliminary polarization parameter reconstruction, and maintain partitioned trigger scanning and local hierarchical early warning release when the link degrades. The main control center is used to complete multimodal fusion, adversarial training and model prediction control, and synchronize data and parameters when communication is restored.

[0015] The warehouse is equipped with an uninterruptible power supply and dual-path mains power switching. An interlock strategy is set between the main control center and the actuator cluster to prevent the risk of condensation and excessive dryness. The main control center stores the graded early warning information and disposal records and establishes batch and storage location association with the warehouse management system for traceability and closed-loop management.

[0016] The beneficial effects of this invention are: This invention achieves proactive sensing, reliable confirmation, and zoned intervention for early mold growth and latent insect activity during tea storage by introducing an acoustic breathing sensing array, a quantum dot gas mapping membrane, an optical excitation imaging subsystem, an infrared polarization interference insect trace detection component, an actuator cluster, and an edge computing unit into the storage and shelving system, with a closed-loop process (S1 to S7) implemented by the main control center. The low-frequency structural vibrations and gas dissipation responses output by the acoustic breathing sensing array under conditions of moisture absorption, release, and slight heating serve as a pre-emptive trigger, directly defining the sectors and timing of optical and polarization scanning. This avoids the energy consumption and redundant data associated with conventional full-domain equal-frequency scanning, allowing the array readout to focus on high-suspicious areas, thereby improving the effective coverage and anomaly detection probability per unit time with the same hardware scale.

[0017] The quantum dot gas mapping film adopts a patch structure consisting of a PET substrate, a sensitive light-emitting layer formed by InP / ZnS quantum dots and molecularly imprinted polymers, and a hydrophobic microporous protective layer. An optical reference area is set up. With constant current pulse width modulation of a narrowband excitation LED array and multi-channel filtering acquisition of an industrial camera, it can stably output a gas concentration distribution heat map with the reference ratio as the core observation quantity in a high-humidity, dusty chamber environment. The sequential sampling of the reference area and the fixed white board provides a benchmark for light decay compensation and aging drift correction. Patch numbering and lifetime counting management reduce calibration and maintenance costs and improve the repeatability of long-term operation.

[0018] The infrared polarization interferometry insect trace detection component outputs a polarization perturbation heatmap under conditions free from visible light interference through long-wave infrared polarization imaging and structural gradient extraction. This heatmap is then spatially registered with the gas concentration distribution heatmap at the main control center, ensuring spatiotemporal co-location of abnormal volatile organic compound emissions before mold growth with insect micro-movement signals. All three modalities participate in evidence accumulation within the consistency scoring framework, effectively reducing false alarms and missed alarms caused by single modalities. The main control center in S5 utilizes a spatiotemporal graph structure to fuse respiratory feature maps, gas concentration distribution heatmaps, and polarization perturbation heatmaps. It also introduces adversarial training using virtual anomaly samples based on a generative model and real operational samples, dynamically updating the consistency scoring threshold and multimodal feature weights. This ensures the scoring strategy remains sensitive and adaptive to unknown anomaly patterns. Simultaneously, it outputs interpretable, graded early warning information including trigger sources and evidence maps, facilitating rapid location and auditable handling by maintenance personnel.

[0019] In S6, the actuator cluster is coordinated and scheduled using model predictive control. This integrates dehumidification setpoints, zoned airflow volume and direction, cold plasma duty cycle, and zoned damper openings into a unified optimization. Under constraints of quality risk and energy consumption, it achieves differentiated intervention for each zone through short-time rolling solutions. Compared to single-loop or threshold-based control, this reduces repeated start-ups and shutdowns and ineffective energy consumption. Combined with zoned isolation and random inspection task assignment, resources are concentrated on risk propagation paths, shortening the closed-loop time from warning to response. In S7, reference self-calibration, acoustic channel health detection, and polarization baseline correction are integrated throughout the entire operation cycle. By compensating for light attenuation in the light source and camera, detecting the amplitude-frequency response and phase consistency of the arrayed MEMS microphone and piezoelectric receiver link, and subtracting the baseline for polarization degree and angle, the impact of environmental changes and device aging on the judgment results is reduced, ensuring the long-term stability of the scoring. The edge computing unit is responsible for the preprocessing and partitioned triggering scheduling of acoustic, optical and polarization data. It maintains local hierarchical early warning release and data buffering during communication degradation. After communication is restored, the main control center completes the data and parameter synchronization and consistency recalculation, avoiding monitoring breakpoints and data loss caused by link fluctuations, and improving the robustness of the system.

[0020] Uninterruptible power supplies (UPS) and dual-path mains power switching provide power redundancy for critical monitoring links and network equipment. Interlocking strategies between the main control center and actuator cluster limit cold source and high-volume direct airflow when dew point risk increases, and restrict dehumidification and cold plasma duty cycles when approaching the lower dryness limit. This ensures that tea moisture content and aroma are not damaged by excessive drying or condensation, reducing quality loss caused by operating condition fluctuations. Tiered early warning information and handling records are linked to batches and storage locations within the warehouse management system. Evidence diagrams, control trajectories, interlock trigger and deactivation times, and model and threshold versions are all included in the same event number traceability link. This makes the entire process of anomaly detection, judgment, handling, and review traceable and verifiable, and allows for rapid tracing of risk expansion paths and upstream causes, supporting the compliance needs of enterprise quality management and regulatory audits.

[0021] By employing acoustic pre-sensory triggering to narrow the scanning range, quantum dot gas mapping to achieve area array visualization, infrared polarization cloaking to confirm life activities, spatiotemporal map structure consistency scoring and adversarial training to maintain sensitivity, model predictive control to reduce energy consumption and improve handling efficiency, reference self-calibration and edge fault tolerance to ensure long-term stability, and traceability closed-loop management, this invention has significant comprehensive technical effects compared to existing warehouse temperature and humidity monitoring and single gas point detection schemes in terms of early warning, false alarm control, energy consumption optimization, operational continuity, and traceability. It can continuously output highly reliable early warnings and achieve refined zoning control and long-term quality maintenance in environments with non-destructive packaging, complex obstruction, and fluctuating humidity and heat. Attached Figure Description

[0022] Figure 1 The process of this invention Figure 1 ; Figure 2 The process of this invention Figure 2 ; Figure 3 The process of this invention Figure 3 . Detailed Implementation

[0023] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] Example 1 like Figure 1 As shown, this embodiment integrates monitoring and execution subsystems and a main control center within a standard warehouse and racking system, forming a closed-loop structure. This structure utilizes an acoustic breathing sensor array for pre-detection triggering, a quantum dot gas mapping film and optical excitation imaging subsystem for area array visualization readout, an infrared polarization interferometry insect trace detection component for concealment confirmation, and an edge computing unit for on-site preprocessing and zoned scheduling. The warehouse is an insulated and moisture-proof structure, and the racking system is arranged in rows and columns. The aisle width accommodates the reciprocating motion of the track-mounted gimbal mechanism. All monitoring nodes and actuator clusters are connected to the main control center and edge computing unit via industrial Ethernet and a power supply bus.

[0025] An acoustic breathing sensing array consists of MEMS microphones and piezoelectric excitation receivers installed at equal intervals along the aisle of the shelving unit. The array elements are mounted on the inner side of the shelving uprights or aisle guardrails, with a spacing of 1-2 meters to cover the cross-section of the aisle. Reference nodes are placed at the corners of the warehouse and at the ventilation ends for background noise modeling. The array channels are connected to the edge computing unit via shielded cables. The sampling frequency is set at 1-2 kHz, with a quantization accuracy of 60 bits. The edge computing unit maintains a unified clock synchronization. Piezoelectric excitation uses low-amplitude sweep frequency or short pulses as structural response excitation. During non-excitation periods, a passive listening mode is maintained to capture spontaneous noise from gas escape. Door opening / closing and forklift passage events are triggered by door magnetic and photoelectric inputs as suppression thresholds. The edge computing unit performs DC removal, bandpass filtering, and transient suppression on the original signal, then performs array beamforming to obtain azimuth estimation. Time-frequency analysis and feature aggregation are then performed at the main control center, outputting abnormal breathing location information including spatial coordinates and anomaly intensity levels. To reduce subsequent imaging energy consumption, the main control center constrains the zonal scanning range, limiting the track-mounted gimbal mechanism to only perform zonal illumination and imaging acquisition within the calibrated sectors.

[0026] The quantum dot gas mapping membrane is applied to the shelf surface and wall areas near cold bridges using a patch structure. Each patch consists of, from bottom to top, a PET substrate layer, a sensitive light-emitting layer formed by cross-linking InP / ZnS quantum dots and molecularly imprinted polymers, and a hydrophobic microporous protective layer. Positioning marks and QR code numbers are located at the four corners of the patch for lifetime counting management. The reference area is formed by locally covering the sensitive light-emitting layer with an inert transparent barrier layer, making this area unresponsive to the target gas and serving as a baseline for optical intensity and aging compensation. Before application, the patches are initially calibrated in a constant temperature and humidity chamber under a standard atmosphere, and the baseline intensity and temperature and humidity coefficients of the reference and sensitive light-emitting areas are recorded in the absence of the target gas. The optical excitation imaging subsystem consists of a narrowband excitation LED array, an industrial camera, a multi-channel filter assembly, and a track-mounted gimbal mechanism. The excitation center wavelength is selected to match the emission characteristics of the quantum dot. The LED array is driven by constant current pulse width modulation to ensure stable illumination. The industrial camera acquires patch images under different filter channels. The track-mounted gimbal mechanism scans along the channel trajectory in a zoned, polling manner. The trigger sequence is issued in real time by the main control center based on abnormal breathing location information to reduce invalid scans. The edge computing unit performs distortion correction, white balance, and background subtraction on the original camera image, separating the reference area and sensitive emitting area of ​​each patch and calculating the intensity ratio. The main control center generates gas signal rasterized data based on this and maps it to the in-cell coordinate system to form a gas concentration distribution heatmap. To enhance repeatability, this embodiment uses the reference ratio as the core observation, defining... Where is the integral intensity of the sensitive emission region in the current frame. This represents the integral intensity of the reference region in the current frame. The main control center calls temperature and humidity sensor data to perform temperature and humidity compensation and uses the factory calibration curve to convert it into a relative concentration grid of the target component, outputting a gas concentration distribution heatmap with timestamps.

[0027] The infrared polarization interferometric insect trace detection component is installed on a shared track with the optical excitation imaging subsystem. It employs a long-wave infrared polarization camera and an infrared window filter structure, covering a defined field of view. Within the same area scanned by the optical system, the main control center synchronously triggers infrared polarization acquisition. The edge computing unit corrects the non-uniformity of the original polarization image and outputs initial values ​​for the polarization parameters. The main control center then resamples the polarization parameters according to the patch coordinates, obtaining a polarization perturbation heatmap of the same size as the gas concentration distribution heatmap for subsequent spatial registration and consistency scoring. To simplify implementation, the polarization heatmap in this embodiment is constructed using the polarization intensity variation rate and the temporal difference of the motion trace, thus meeting the requirements for trace confirmation without the need for complex multi-frame interferometric reconstruction.

[0028] Data processing operates under a dual mechanism of fixed-cycle and event-triggered processing. During the fixed-cycle period, the main control center schedules polling each channel at a lower frequency when no abnormal respiratory location information is available. Event-triggered processing immediately narrows the scanning range and increases the imaging frequency of the corresponding sector when abnormal respiratory location information is detected. The main control center performs spatial registration of the respiratory feature map, gas concentration distribution heatmap, and polarization perturbation heatmap within the same time window. The registration coordinate system is jointly determined by the mileage encoding of the orbital gimbal mechanism and the QR code positioning on the patch. Errors are locally corrected using reference markers. To form interpretable early warning judgments, this embodiment employs a consistency scoring strategy, summing the three types of normalized anomalies according to their weights to obtain a consistency score (S), defined as follows: Where (A) is the normalized anomaly intensity of the corresponding sector in the respiratory feature map, (G) is the normalized anomaly intensity of the corresponding grid in the gas concentration distribution heatmap, and (P) is the normalized anomaly intensity of the corresponding grid in the polarization perturbation heatmap. These are the weighting coefficients determined during factory calibration and initial operation. The main control center outputs graded early warning information based on the comparison between (S) and the threshold. A mild warning corresponds to (S) exceeding the warning threshold and gas anomalies being the primary cause; a moderate warning corresponds to (S) exceeding a higher threshold and gas and polarization anomalies occurring simultaneously; and a severe warning corresponds to (S) continuously exceeding the highest threshold and showing an increasing trend.

[0029] The execution layer is controlled by a cluster of actuators responding to tiered early warning information to implement zoned regulation. The variable frequency dehumidification unit receives fine-tuning commands for the dehumidification setpoint to maintain the zoned dew point below the set upper limit. The zoned air supply unit forms directional airflow within the same channel according to the flow direction to break up the accumulated moisture microenvironment. The cold plasma sterilization unit does not operate under mild warnings, performs spot sterilization with a low duty cycle under moderate warnings, and continues to operate within safety limits under severe warnings. The zoned damper unit adjusts its opening based on the upstream and downstream pressure difference strategy of the protected channel to form a micro-pressure difference. Each execution action has a limited minimum execution duration and cooling interval to avoid frequent start-stop. The main control center records all execution parameters and corresponding warning levels, timestamps, and spatial locations as reference data for the re-monitoring phase. During the re-monitoring phase, the same sector undergoes rapid acoustic scanning and a small amount of optical inspection at a higher frequency. If (S) falls below the reset threshold within two consecutive windows, the zoned execution intensity is reduced and regular polling is resumed; if (S) does not decrease, the execution layer action is maintained or upgraded.

[0030] To ensure the long-term stability of the array readout, this embodiment automatically initiates a reference self-calibration process at the start of each shift. The track-mounted gimbal mechanism first acquires optical baselines at fixed whiteboard positions in each channel for compensation of light attenuation in excitation light and camera sensitivity. Then, it briefly illuminates the reference area of ​​each patch to calculate the reference ratio for the shift and generates a lifetime count report. When the reference ratio falls below the maintenance threshold, a replacement prompt is triggered on the monitoring interface, and the patch is marked as a priority for inspection. On the acoustic link, the edge computing unit detects the health status of each array element channel through low-amplitude frequency sweeping, marks abnormal channels, and prompts for maintenance on the main control interface. The aforementioned self-calibration and health detection information participates in the subsequent consistency scoring weight adjustment, ensuring that the score is not misled by drift in a single sensing link.

[0031] The system integration method in this embodiment is an edge-first architecture, where edge computing units handle acoustic, optical, and polarization preprocessing and partitioned triggering scheduling, while the main control center completes spatial registration, consistency scoring, hierarchical early warning, and generation of partitioned control instructions. All data is stored in a local database with timestamps and spatial coordinate indexes, generating gas concentration distribution heatmaps, polarization perturbation heatmaps, and respiration characteristics. Figure 3 The visualization results are output to the monitoring interface. According to this process, under typical humid weather and high-stack storage conditions, the system can quickly locate abnormal sectors with an acoustic breathing sensing array, generate a thermal map of area gas concentration distribution with reference compensation using a quantum dot gas mapping film and an optical excitation imaging subsystem, confirm the presence of hidden traces with an infrared polarization interferometry insect trace detection component, output graded early warning information through consistency scoring and drive the actuator cluster to implement low-energy partition control, and finally achieve closed-loop convergence with reference self-calibration and re-monitoring, realizing the advanced perception and handling of the risk of mold and moisture accumulation in tea.

[0032] Example 2 like Figure 1 and Figure 2 As shown, this embodiment, based on the hardware and deployment of Embodiment 1, introduces a deep processing flow of the infrared polarization interferometric insect trace detection component, a consistency scoring and adversarial learning mechanism for the spatiotemporal graph structure, and a model predictive control strategy for the actuator cluster, forming an enhanced closed loop of acoustic premonition triggering—quantum dot gas mapping membrane array readout—infrared polarization concealment confirmation—consistency threshold determination—cooperative regulation—parameter refeedback. Components and steps identical to those in Embodiment 1 are only briefly described; the focus is on the newly added and enhanced parts, ensuring consistency in terminology and functional chains.

[0033] The infrared polarization interferometric insect trace detection component employs a long-wave infrared polarization camera and an infrared window filter, mounted on a track-mounted gimbal mechanism shared with the optical excitation imaging subsystem. The line of sight is aligned with the normal of the quantum dot gas mapping film patch using a tooling fixture. After each abnormal breathing location information output by the acoustic breathing sensing array, the main control center limits the scanning range to the calibrated sector. The track-mounted gimbal mechanism first completes the partitioned illumination of the quantum dot gas mapping film and imaging by the industrial camera, then initiates infrared polarization acquisition in the same sector, ensuring that the two types of imaging data are approximately co-located in time and space. The edge computing unit performs non-uniformity correction, bad pixel interpolation, and blackbody short reference calibration at the imaging end, outputting a sequence of original polarization images and a baseline of thermal radiation intensity. The main control center performs multi-frame fusion on this sequence to suppress random noise and micro-jitter, employing a weighted median and optical flow consistency constraint fusion strategy using a sliding time window to obtain the fused intensity sequence for that sector. Polarization parameter reconstruction is based on the multi-orientation sub-pixel intensity output by the camera, reconstructing Stokes parameters and calculating the degree of polarization and polarization angle to obtain a polarization parameter grid. The degree of polarization (DoLP) and the polarization angle (AoLP) are the fundamental observations of this component, and their calculation is as follows: Based on the fused polarization parameters, the main control center calculates the structural gradient and performs morphological constraint-based connected component screening. Grids with persistent micro-motion characteristics within high-gradient connected components are marked as candidate insect trace regions. Combined with the slightly elevated background profile of thermal radiation intensity, a normalized polarization perturbation heatmap is generated. To achieve spatial registration with the gas concentration distribution heatmap, the system uses short-time infrared spot illumination on the positioning markers at the four corners of the patch for coordinate correction. Furthermore, it employs a dual index of orbital mileage and patch QR codes to resample the polarization perturbation heatmap to match the gas concentration distribution heatmap. Figure 1 Consistent raster resolution.

[0034] Multimodal consistency determination employs a spatiotemporal graph-structured data fusion model. The central control unit maps the respiratory feature map, gas concentration distribution heatmap, and polarization perturbation heatmap obtained from the scanned sector within the current time window into graph-structured node features. Nodes are defined as patch grid cells, and edge weights are the weighted sum of the ventilation connectivity and historical covariance of adjacent grids, forming a spatiotemporal topology. The consistency score at the current time is calculated on this topology. Considering temporal stability and the influence of historical states, a decay term from the previous time step's score is introduced to obtain the consistency score of grid (i) at time (t). in( The normalized anomaly intensity of the respiratory feature map in grid (i), The normalized anomaly intensity of the gas concentration distribution heatmap at grid (i), The normalized anomaly intensity of the polarization perturbation heatmap at grid (i) is ( For three-modal weights, This is the time smoothing coefficient, with a range of values. The scoring system performs threshold comparisons at the central control unit and outputs tiered early warning information. Simultaneously, it generates an evidence map, which consists of a visualization layer overlaid with the three-modal anomaly intensity and the scoring, serving as a basis for manual review and record-keeping of actions.

[0035] To maintain sensitivity to unknown anomalies, the system employs an adversarial training process based on a generative model, using virtual anomaly samples and real-world running samples. Virtual anomaly samples are synthesized offline by the central control unit during idle windows that do not affect real-time decision-making. The synthesis method involves injecting a thin VOC plume with a controllable diffusion nucleus into the gas concentration distribution heatmap and superimposing a micro-motion vector field satisfying boundary conditions onto the polarization perturbation heatmap. Simultaneously, the phase relationship and energy upper limit with the respiration feature map are constrained to ensure the physical plausibility of the virtual samples. The central control unit aggregates positive and negative sample sets from real-world running samples using a sliding time window, combining them with virtual samples to form training batches. A discriminator updates the consistency scoring threshold and modal weights, and early stopping is achieved by using false alarm rate and false negative rate indicators on the verification set to avoid threshold oscillations. After training, the threshold and weights are written back to the online scoring process in a version-controlled manner, enabling adaptive updates to the consistency scoring strategy. Simultaneously, the interpretable content of the evidence map is recorded and archived with the model version, meeting quality management and auditing requirements.

[0036] The coordinated scheduling of the actuator cluster employs a model predictive control strategy. Control variables include the dehumidification setpoint, airflow volume and direction, cold plasma duty cycle, and zone damper opening. The prediction time domain is short-term finite, and the sampling and rolling optimization cycles are aligned with the scan cycle. The cost function includes quality risk terms, energy consumption terms, and execution smoothing terms, forming... in( The risk indicator for the predicted time is obtained by weighted integration of the consistency score grid field over the controlled partition. For energy consumption estimation, calculations are performed under given operating conditions using power models for dehumidification, air supply, and cold plasma. The change range of the control quantity between two adjacent steps. The constraint coefficients are used for trade-offs. Constraints include: zone dew point not exceeding the upper limit; ozone concentration generated by cold plasma not exceeding the safety threshold; air supply volume and zone damper opening within the equipment's rated range; minimum start / stop time for the dehumidifier; and cooling interval. The solver outputs the control quantity for each rolling step. The main control center sends the control quantity to the actuator cluster and generates a disposal order. In severe warning conditions, additional zone isolation and sampling task orders are added. The sampling task order includes the location identifier, sampling quantity, sampling time limit, and delivery process. After disposal, the system retrieves the test results and stores them in association with the evidence map.

[0037] During operation, the central control unit monitors the spatial continuity of the scoring results. If the scoring field shows a continuous increase along the main ventilation path in adjacent grids, and the gradient direction is consistent with the guide direction, the path is marked as a risk expansion path. Priority is given to allocating adjustment resources for the air supply guide direction and the opening of the zone dampers. If necessary, cold plasma low duty cycle protection is pre-deployed in the upstream sector to reduce the probability of mold or pests spreading along the airflow. Dynamic balance of energy consumption is achieved through the adjustment of energy consumption upper limit and peak and valley price periods. During peak price periods, the weight of risk items is appropriately increased to ensure quality priority; during valley price periods, the intensity of active intervention is increased to improve the completion of drying and disinfection. All control and judgment parameters are archived with timestamps, zone numbers, and model version numbers. The real-time interface displays the scoring field, evidence diagram, control quantity trajectory, and energy consumption curve, and provides an explanation of the three-modal contribution rate for each judgment, facilitating rapid review by the operations team.

[0038] The new process in this embodiment is implemented on the same hardware architecture as in Embodiment 1, without changing the deployment of the acoustic breathing sensing array and the quantum dot gas mapping membrane. Only infrared polarization acquisition and polarization parameter reconstruction are inserted into the scheduling sequence of the track-type gimbal mechanism. At the data level, the reliability of consistency scoring is enhanced through spatiotemporal graph structure and adversarial training. At the control level, model predictive control is introduced to achieve the dual-objective optimization of quality risk and energy consumption cost. At the end of each closed loop, the system feeds back the updated results of the scoring threshold, modal weights, and actuator response coefficients to the next cycle, forming an adaptive evolution mechanism from perception to decision-making to regulation and parameter updates. Through the above structured implementation, this embodiment can stably output highly consistent fusion results of polarization disturbance heatmap, gas concentration distribution heatmap, and breathing feature map under conditions of complex obstruction, variable airflow, and load fluctuations within the storage facility. This significantly reduces the false alarm rate and increases the amount of early warning. Simultaneously, within a controllable energy consumption range, it effectively inhibits zoned dew point and microbial activity, meeting the dual requirements of quality and stability for long-term tea storage.

[0039] Example 3 like Figures 1 to 3 As shown, this embodiment, based on Embodiments 1 and 2, is designed for continuous availability under long-term engineering operation and abnormal scenarios. It constructs an integrated mechanism for self-inspection and self-calibration, edge fault tolerance, and power supply and traceability, ensuring the monitoring and execution subsystem maintains consistent execution of partitioned trigger scanning, local hierarchical early warning issuance, and safety interlock control strategies even under conditions of communication degradation, power fluctuations, and device aging. The layout and signal flow of the warehouse, racking system, acoustic breathing sensing array, quantum dot gas mapping membrane, optical excitation imaging subsystem, infrared polarization interferometry insect trace detection component, actuator cluster and main control center, and edge computing unit are consistent with the aforementioned embodiments and will not be repeated. Only the engineering details not covered in Embodiments 1 and 2 are disclosed.

[0040] The system automatically enters the self-test and self-calibration process during the shift switching window corresponding to S7. The optical self-test is performed by the track-type gimbal mechanism, which sequentially positions the fixed whiteboard and the reference area of ​​each patch. The narrowband excitation LED array outputs the nominal illuminance step using constant current pulse width modulation. The industrial camera acquires baseline frames under the multi-channel filter assembly. The main control center generates the light source and camera light decay compensation coefficient and dark current correction coefficient based on this and writes them back to the calibration table of the optical excitation imaging subsystem. Subsequently, the reference area of ​​each quantum dot gas mapping film is briefly illuminated and acquired. After the edge computing unit completes distortion correction and region segmentation, it outputs the reference intensity. The main control center compares the reference intensity with the factory calibration value, updates the patch life count, and marks the patch into the priority inspection list when it is below the maintenance threshold. During the online scanning process, the modal weight of the patch in the consistency score is reduced accordingly to avoid misjudgment caused by aging. The acoustic self-test uses an edge computing unit to issue low-amplitude frequency sweep excitation during non-production periods, reading the amplitude-frequency response and channel phase consistency of the array MEMS microphone and piezoelectric receiver link. The main control center calculates the signal-to-noise ratio, crosstalk index, and phase deviation of each array element, generating an acoustic channel health status matrix and marking abnormal channels. Abnormal channels are automatically included in channel shielding or weight attenuation during subsequent array beamforming in S1 until the marking is turned off during maintenance. The polarization self-test uses a long-wave infrared polarization camera to acquire a reference frame series from a high-emissivity blackbody panel. The main control center performs multi-frame fusion on this frame series and calculates the baseline offset of polarization parameters, obtaining the baseline correction amount of polarization degree and polarization angle. It updates the correction table of the infrared polarization interferometric insect trace detection component and performs baseline subtraction during online imaging to ensure the comparability of polarization perturbation heatmaps. When an abnormality is detected at any stage of the self-test process, a fault alarm is immediately triggered. The alarm content includes component identification, abnormal indicators, suggested handling, and whether it affects the real-time score. The main control center simultaneously generates a maintenance work order on the monitoring interface and pushes it to the on-duty personnel.

[0041] Edge fault tolerance achieves adaptive degradation through heartbeat detection and latency measurement between the edge computing unit and the main control center. The main control center periodically issues clock synchronization packets and scheduling instructions, and the edge computing unit records round-trip latency and packet loss rate. When the latency and packet loss exceed the threshold for consecutive periods, it enters a communication degradation state. The edge computing unit switches to the local scheduling table to perform partition-triggered scanning, retains the acoustic preprocessing, preliminary optical image calculation, and preliminary polarization parameter reconstruction from Embodiments 1 and 2, and independently outputs local graded early warning information to the on-site audible and visual alarms. At the same time, the index summary and key features of the original data are written to the local circular buffer in chronological order to prevent data loss due to power failure. After communication is restored, the master control center initiates a data and parameter synchronization handshake. First, it retrieves the partition trigger logs and local hierarchical early warning information from the edge. Then, it requests data fragments in order of timestamp and partition number. The edge computing unit verifies and returns the data fragments. After the master control center completes gap filling, it performs consistency recalculation and evidence graph reconstruction. If the edge has made temporary adjustments to modal weights or thresholds, the master control center resolves conflicts based on the model version and retains records of the edge adjustments for future reference, ensuring score traceability and result uniqueness. Throughout the fault tolerance process, control of the actuator cluster is not delegated to the edge to avoid inconsistent actions with the master control center during communication degradation. Only local hierarchical early warning issuance and passive triggering of safety interlocks are allowed on-site. After communication is restored, the master control center uniformly issues partition control instructions and handling work orders.

[0042] Power supply and safety interlocks are guaranteed by an uninterruptible power supply (UPS) and a dual-channel AC power switching device, forming a three-channel power supply architecture consisting of primary AC power, secondary AC power, and a UPS. The UPS connects to the main control center, edge computing unit, acoustic breathing sensor array, minimum monitoring loads of industrial cameras and long-wave infrared polarization cameras, and network switching unit. The actuator cluster connects to the dual AC power supply and uses a switching device for seamless switching. The main control center collects the voltage, frequency, and switching status of the three power supplies in real time. When AC power fails, it quickly enters a load grading strategy, first ensuring the continued operation of the acoustic breathing sensor array, network switching unit, and edge computing unit. Then, it reduces the scanning frequency of the optical excitation imaging subsystem and the infrared polarization interference insect trace detection component, retaining the inspection of calibrated sectors. Finally, it completes necessary data write-to-disk and log archiving within the UPS time limit. When power is restored, the system returns to normal scanning frequency using a fixed whiteboard and reference area rapid calibration process. To prevent the risks of condensation and excessive drying, an interlock strategy is configured between the main control center and the actuator cluster. The main control center calculates the dew point risk of each zone based on the zone's temperature, humidity, and surface temperature. When the dew point risk exceeds the threshold, the cold source or large airflow is prohibited from directly impacting the target surface. Instead, a combination of dehumidification and gentle airflow is used until the dew point risk is eliminated. The interlock process is recorded in the control trajectory and displayed as an interlock status bar on the monitoring interface. At the same time, a lower limit is set for excessive drying. If the relative humidity of a zone is close to the lower limit, the main control center restricts the increase of the dehumidification setpoint and the cold plasma duty cycle to ensure that the moisture content and aroma of the tea are not damaged.

[0043] The system's traceability function spans the entire management chain from early warning to handling and review. The central control unit writes tiered early warning information and handling records into the database using a dual index of batch number and location code, and establishes a foreign key relationship with the data tables of the warehouse management system. Early warning events are associated with evidence map files such as respiratory feature maps, gas concentration distribution heatmaps, and polarization disturbance heatmaps, the current actuator control quantity and its time trajectory, the time points of interlock triggering and de-interlocking, model version number, and scoring threshold version number. Detection results backfilled after sampling are merged with the same event number, forming a closed-loop record. When cross-zone expansion occurs, the central control unit constructs a risk path map based on the control history of zone damper opening and airflow direction. The traceability page displays the expansion path and corresponding evidence maps for auditing and quality review. For maintenance closure, anomalies generated by the central control unit during the self-inspection phase are automatically converted into maintenance work orders, including component identification, abnormal indicators, suggested replacement cycle, and shutdown window. After the work order is completed, the system checks whether the health matrix has returned to the green zone; if not, it continues to alarm and suggests further investigation.

[0044] Data processing remains continuous and uninterrupted in fault-tolerant scenarios. Edge computing units continuously perform acoustic, optical, and polarization preprocessing and partitioned triggering scheduling. During communication degradation, local graded early warning information is released via audible and visual alarms to ensure on-site response. After communication is restored, the main control center completes the unified convergence of multimodal fusion, adversarial training, and model predictive control, ensuring consistency between the scoring and execution strategies. On the output side, the system maintains the availability of early warning information, evidence maps, and control trajectories under any power and communication conditions. In degraded states, the rendering and uploading frequency of non-critical layers is reduced, prioritizing the updating of evidence maps and early warning broadcasts for the identified sectors. In the long-term operation dimension, the main control center gradually updates the health weights and threshold versions of each mode based on the self-inspection results and maintenance work orders in a closed loop. It also establishes a lifetime prediction curve for the reference ratio decay trend of the quantum dot gas mapping membrane and includes patches that are about to reach the maintenance threshold in the "key polling list" of the next shift. Through the engineering path of "preemptive triggering - array readout - hidden trace confirmation - consistency judgment - collaborative control - self-inspection and self-calibration - fault tolerance synchronization - traceability closed loop", it ensures that the warehouse can still achieve stable closed-loop operation of monitoring, decision-making, control and re-monitoring under abnormal and fluctuating conditions.

[0045] Example 4 This embodiment quantitatively evaluates the contribution of the key technical features of the present invention to indicators such as monitoring accuracy, anomaly detection rate, energy consumption, response time, early warning lead time, and system availability through comparative and ablation experiments. It also verifies the acoustic breathing sensing array, quantum dot gas mapping film, polarization cloaking confirmation, and spatiotemporal... Figure 1 The effectiveness and necessity of technical aspects such as consistency scoring, model predictive control, self-checking and self-calibration, and edge tolerance were examined. The experiment was conducted based on the architecture and layout schemes of Examples 1, 2, and 3, under the same warehouse, the same shelf layout, and the same tea stacking method. The experimental period was thirty days, with two manual spot checks per day as external references.

[0046] To verify the advantages of this invention over traditional technologies, a traditional temperature and humidity sensing and dehumidification control scheme was selected as a control group. Comparative tests were conducted on indicators such as monitoring accuracy, anomaly detection rate, energy consumption, execution response time, early warning lead time, false alarm rate, missed alarm rate, array coverage, and effective online duration. The control group used point-type temperature and humidity sensors arranged in rows and columns; when the monitored value exceeded a preset threshold, the dehumidification or humidification unit was triggered.

[0047] The results of the comparative experiment are shown in the table below: Example 1 improved monitoring accuracy by approximately 11 percentage points and anomaly detection rate by 14 percentage points compared to the control group, mainly due to acoustic premonition triggering and the spatial resolution capability of the quantum dot film array; Example 2 improved monitoring accuracy by introducing polarization trace confirmation and spatiotemporal... Figure 1Consistency scoring and anomaly detection rate were further improved to 95%; in Example 3, after adding self-checking and edge fault tolerance, the anomaly detection rate reached 96%, and the system robustness was significantly enhanced. Example 1 reduced energy consumption by about 29% compared to the control group, mainly by achieving on-demand scanning through acoustic pre-sensory triggering, avoiding full-domain polling; Example 2, due to the addition of a long-wave infrared polarization camera and MPC rolling optimization calculation, saw a slight increase in energy consumption to 11.2 kWh per day, but through MPC multi-objective optimization, it was still better than the control group while ensuring quality; Example 3, due to the addition of a self-checking process and UPS load protection, saw a further increase in energy consumption to 11.5 kWh per day, but the effective online time increased to 99.3%, and the overall energy efficiency ratio was still better than the traditional solution. Example 1 shortened the response time to 5.2 minutes through acoustic positioning, an improvement of about 39% compared to the control group; Example 2's MPC rolling optimization further shortened it to 4.8 minutes; Example 3, through local preprocessing and partitioned trigger scheduling of the edge computing unit, reduced the response time to 4.6 minutes. Example 1 achieved a warning lead time of 16.8 hours, four times that of the control group. Example 2, through multimodal consistency scoring and adversarial learning, increased the warning lead time to 20.5 hours, reducing the false alarm rate to 4.9% and the missed alarm rate to 6.1%. Example 3, through self-checking and self-calibration to ensure the health of the sensor link, further increased the warning lead time to 21.8 hours, reducing the false alarm rate to 4.2% and the missed alarm rate to 5.4%. The traditional temperature and humidity solution has an array coverage of only 22%, limited by the spatial resolution of point sensors. Example 1 increased the coverage to 65% through quantum dot film patch deployment. Example 2 increased it to 69% through synchronous scanning strategy of polarization components. Example 3 increased the coverage to 71% through self-checking and self-calibration for priority polling of aging patches and degradation protection of edge fault tolerance, while achieving an effective online time of 99.3%, significantly better than the control group's 97.1%.

[0048] To quantitatively evaluate the contribution of each key technical feature to the overall system performance, this embodiment further conducts ablation experiments. Using the complete architecture of Embodiment 3 as a baseline, this architecture includes an acoustic breathing sensing array, a quantum dot gas mapping film, an optical excitation imaging subsystem, polarization tracing confirmation, and a spatiotemporal... Figure 1 Consistency scoring, MPC collaborative scheduling, self-checking and self-correction, and edge tolerance were implemented. Specific technical steps were removed one by one, and the degradation of performance indicators was observed. The ablation group included: removing polarization occultation confirmation while retaining only the acoustic and gas dual modes, and removing spatiotemporal... Figure 1 The ablation test results were improved by replacing consistency scoring with fixed-weight simple weighting, removing MPC collaborative scheduling and replacing it with traditional PID independent control, and eliminating edge fault tolerance and self-checking and self-correction, with centralized processing in the main control center and no degradation protection. The ablation test results are shown in the table below: After removing the long-wave infrared polarization component, the anomaly detection rate decreased from 96% to 93%, while the false alarm rate increased from 5.4% to 7.5%, indicating that polarization trace confirmation plays an irreplaceable role in detecting subtle anomalies such as insect pests. Simultaneously, energy consumption decreased to 10.9 kWh per day, verifying that the additional energy consumption of the polarization camera and infrared window filter is approximately 0.6 kWh per day. Removing the spatiotemporal graph structure fusion and adversarial learning mechanisms, and replacing them with a simple weighted summation with fixed weights, reduced the monitoring accuracy to 92%, the anomaly detection rate to 91%, the false alarm rate to 7.8%, the false alarm rate to 8.9%, and the early warning lead time to 17.5 hours; indicating that the spatiotemporal graph structure fusion and adversarial learning mechanisms play an irreplaceable role in detecting subtle anomalies such as insect pests. Figure 1 Consistency scoring and dynamic threshold updates are crucial for reducing false alarms and missed alarms, and improving early warning sensitivity. Energy consumption increased slightly to 11.8 kWh per day due to frequent actuator start-stop caused by false triggering. After removing model predictive control and replacing the actuator cluster with traditional PID independent control, energy consumption increased significantly to 13.1 kWh per day, and the execution response time increased to 6.2 minutes, indicating that MPC's multi-objective rolling optimization and execution smoothing constraints have significant advantages in reducing energy consumption and shortening response time. Early warning lead time decreased to 18.3 hours, but both false alarm rate and missed alarm rate increased. After removing the degradation fault tolerance mechanism of the edge computing unit and the S7 self-check and self-calibration process, the effective online time decreased from 99.3% to 97.8%, verifying the contribution of edge fault tolerance and self-checking to the system's high availability. Monitoring accuracy and anomaly recognition rate decreased slightly due to the lack of real-time compensation for sensor link aging. Energy consumption decreased to 11.3 kWh per day because the optical baseline acquisition and frequency sweep excitation of the self-checking process were omitted.

[0049] The above description is merely a preferred embodiment of the present invention. It should be understood that the present invention is not limited to the forms disclosed herein and should not be construed as excluding other embodiments. It can be used in various other combinations, modifications, and environments, and can be altered within the scope of the concept described herein through the above teachings or related technologies or knowledge. Modifications and variations made by those skilled in the art that do not depart from the spirit and scope of the present invention should be within the protection scope of the appended claims.

Claims

1. A smart monitoring and early warning warehouse for tea storage, comprising a warehouse body, a shelving system, a monitoring and execution subsystem, and a main control center, characterized in that: The monitoring and execution subsystem includes an acoustic respiration sensing array, a quantum dot gas mapping membrane, an optical excitation imaging subsystem, an infrared polarization interferometry insect trace detection component, an actuator cluster, and an edge computing unit. The acoustic respiration sensing array is installed in the shelf aisles and corner areas of the warehouse. It includes an array of MEMS microphones and piezoelectric excitation receivers to collect low-frequency structural vibrations and gas emission acoustic responses generated by the tea pile under conditions of moisture absorption, moisture release, and internal micro-heating, and outputs abnormal respiration location information. The quantum dot gas mapping membrane is applied in a patch-like manner to the shelf surface and the wall area near the cold bridge. The quantum dot gas mapping membrane includes, from bottom to top, a PET substrate layer and a sensitive luminescent layer formed by InP / ZnS quantum dots and molecularly imprinted polymers. The quantum dot gas mapping film comprises a layer and a hydrophobic microporous protective layer. An optical reference area is embedded within the quantum dot gas mapping film and managed using patch numbering. The optical excitation imaging subsystem includes a narrowband excitation LED array, an industrial camera, a multi-channel filter assembly, and a track-type gimbal mechanism, used to perform partitioned polling scanning and acquire optical images of the area covered by the quantum dot gas mapping film. The infrared polarization interferometry insect trace detection assembly includes a long-wave infrared polarization camera and an infrared window filter, used to acquire polarization parameters and output polarization perturbation thermal images. The actuator cluster includes a frequency conversion dehumidification unit, a partitioned air supply unit, a cold plasma sterilization unit, and a partitioned damper unit. The edge computing unit is communicatively connected to the main control center, and is used to preprocess acoustic, optical, and polarization raw data and perform partitioned trigger scheduling. The main control center is configured to perform the following steps: S1: Based on the array beamforming and time-frequency analysis of the acoustic breathing sensing array, generate a breathing feature map and output abnormal breathing location information; S2: Based on the abnormal breathing location information, control the optical excitation imaging subsystem to perform partitioned illumination and imaging acquisition of the corresponding covered area, and limit the scanning range to the sector marked by the abnormal breathing location information; S3: Optically compare the reference region and the sensitive light-emitting region of the quantum dot gas mapping film to generate a heat map of the gas concentration distribution of volatile organic compounds and ammonia; S4: During the partitioned polling scan, the infrared polarization interference insect trace detection component is synchronously triggered to image the corresponding area, calculate the polarization parameters, and generate a polarization perturbation heat map; S5: Spatial registration and consistency scoring are performed on the respiratory feature map, the gas concentration distribution heat map and the polarization perturbation heat map. Based on the determination result that the consistency score exceeds the threshold, graded early warning information is output. S6: Based on the hierarchical early warning information, a zone control command is sent to the actuator cluster to complete the coordinated scheduling of dehumidification setpoint, air supply volume and flow direction, cold plasma duty cycle and zone damper opening. S7: Based on the execution results and monitoring data, update the scanning frequency, excitation power, polarization sampling beat and acoustic excitation amplitude, and perform reference self-correction on the optical attenuation of the quantum dot gas mapping film, forming a closed-loop operation of monitoring, decision-making, regulation and re-monitoring.

2. The intelligent monitoring and early warning warehouse for tea storage according to claim 1, characterized in that: The acoustic breathing sensing array is deployed at equal intervals along the direction of the shelf aisle, and reference nodes are set at the corners of the warehouse and the ventilation end for background noise modeling. In step S1, the main control center performs door opening and closing event elimination, transient suppression and beam pointing constraint, and outputs abnormal breathing location information including spatial positioning coordinates and abnormal intensity level.

3. The intelligent monitoring and early warning warehouse for tea storage according to claim 1, characterized in that: The sensitive light-emitting layer of the quantum dot gas mapping membrane is a composite membrane formed by cross-linking InP / ZnS quantum dots that have been stabilized by ligands with molecularly imprinted polymers for ethanol and ammonia. The hydrophobic microporous protective layer covers the composite membrane and retains gas diffusion channels. The main control center records the patch number and lifetime count and triggers a replacement prompt when the reference ratio is lower than the threshold.

4. The intelligent monitoring and early warning warehouse for tea storage according to claim 1, characterized in that: The narrowband excitation LED array of the optical excitation imaging subsystem is driven by constant current pulse width modulation and operates according to a zoned lighting strategy. The industrial camera is equipped with a multi-channel filter component. In step S3, the main control center performs temperature and humidity compensation based on the ratio of the intensity of the reference region to the intensity of the sensitive luminescent region, and outputs a calibrated gas concentration distribution heat map.

5. The intelligent monitoring and early warning warehouse for tea storage according to claim 1, characterized in that: The infrared polarization interferometric insect trace detection component acquires the original polarization image through a long-wave infrared polarization camera. In step S4, the main control center performs multi-frame fusion, polarization parameter reconstruction and structural gradient extraction, and spatially registers it with the gas concentration distribution heat map obtained in step S3 to output a polarization perturbation heat map.

6. The intelligent monitoring and early warning warehouse for tea storage according to claim 1, characterized in that: In step S5, the main control center uses a spatiotemporal graph structure data fusion model to determine the consistency of the respiratory feature map, the gas concentration distribution heatmap, and the polarization perturbation heatmap. It also introduces virtual abnormal samples based on the generative model to conduct adversarial training with real running samples, dynamically updates the consistency score threshold and multimodal feature weights, and outputs interpretable hierarchical early warning information containing the trigger source and evidence map.

7. The intelligent monitoring and early warning warehouse for tea storage according to claim 1, characterized in that: The actuator cluster is coordinated and scheduled by the main control center according to the model predictive control strategy. In step S6, with quality risk and energy consumption as the target constraints, the linkage control quantities of dehumidification set point, air supply volume and flow direction, cold plasma duty cycle and partition damper opening are generated. In the state of severe warning, the partition isolation and spot check task order are assigned.

8. The intelligent monitoring and early warning warehouse for tea storage according to claim 1, characterized in that: In step S7, the main control center performs optical self-test, acoustic self-test, and polarization self-test. The optical self-test includes sequential sampling of the reference area of ​​the quantum dot gas mapping film and the fixed white board and compensation for light attenuation of the light source and camera. The acoustic self-test includes low-amplitude frequency sweep detection of the health status of the array channel. The polarization self-test includes polarization baseline correction. When an abnormality is detected, a fault alarm is triggered.

9. The intelligent monitoring and early warning warehouse for tea storage according to claim 1, characterized in that: The edge computing unit is used to complete acoustic signal preprocessing, preliminary optical image calculation and preliminary polarization parameter reconstruction, and maintain partitioned triggered scanning and local hierarchical early warning release when the link degrades. The main control center is used to complete multimodal fusion, adversarial training and model prediction control, and synchronize data and parameters when communication is restored.

10. The intelligent monitoring and early warning warehouse for tea storage according to claim 1, characterized in that: The warehouse is equipped with an uninterruptible power supply and dual-path mains power switching. An interlock strategy is set between the main control center and the actuator cluster to prevent the risk of condensation and excessive drying. The main control center stores graded early warning information and disposal records and establishes batch and storage location associations with the warehouse management system for traceability and closed-loop management.

Images