Reduced-pressure fumigation fresh-keeping device and method for fruits

Abstract

The invention discloses a fruit decompression fumigation fresh-keeping device and method. The device comprises a sealed chamber; the pressure reduction unit is connected with the sealing cavity through a first pipeline, and the pressure reduction unit comprises a vacuum pump and a pressure sensor; the fumigation unit is connected with the sealing chamber through a second pipeline, and the fumigation unit comprises a fumigation gas storage tank and a gas mass flow controller; the environment regulation and control unit comprises a refrigerating machine connected with the sealing chamber through a third pipeline, a humidifier connected with the sealing chamber through a fourth pipeline, a temperature sensor and a humidity sensor; and the control unit is used for storing a fresh-keeping parameter prediction model trained based on a machine learning algorithm, and outputting a target temperature value, a target humidity value, a target pressure value and fumigation maintenance time by taking the fruit type information, the initial quality parameters and the target storage period as inputs. The storage period and shelf life of the fruits can be remarkably prolonged, and the initial hardness, flavor and nutritional ingredients of the fruits are effectively maintained.

Description

Technical Field

[0001] This invention relates to the technical field of fruit preservation. More specifically, this invention relates to a fruit decompression fumigation preservation device and method. Background Technology

[0002] Fruits such as peaches, plums, cherries, and apricots hold an important position in the market due to their tender, juicy flesh, unique flavor, and rich nutrition. However, these fruits have high physiological activity and respiration rate after harvest, and their cell walls are easily degraded, leading to softening of the flesh. They are also susceptible to infection by pathogens, causing rot. Furthermore, their thin skin and high water content make them prone to wilting due to water loss, severely shortening their storage period and shelf life.

[0003] Among existing preservation technologies, cold storage can only reduce the metabolic rate and is difficult to effectively inhibit the growth of pathogens. Long-term storage can easily lead to chilling injury and quality deterioration. Ordinary fumigation preservation has defects such as poor control of gas concentration and inability to adapt to the characteristics of different varieties of stone fruits. Simple decompression preservation lacks precise and coordinated control of temperature and humidity, and it is difficult to take into account both water loss prevention and quality maintenance.

[0004] With the increasing demand in the fresh food market for year-round fruit supply, long-distance transportation, and quality stability, existing technologies are insufficient to achieve comprehensive optimization of the storage environment, and a technical solution that can overcome the above-mentioned shortcomings is urgently needed. Summary of the Invention

[0005] One objective of this invention is to provide a fruit decompression fumigation preservation device and method that can significantly extend the storage period and shelf life of fruits, and effectively maintain the initial firmness, flavor and nutritional components of the fruits.

[0006] To achieve these objectives and other advantages of the present invention, according to one aspect of the present invention, a fruit depressurization fumigation preservation device is provided, comprising: a sealed chamber for containing fruit; a depressurization unit connected to the sealed chamber via a first pipeline, the depressurization unit including a vacuum pump and a pressure sensor for detecting the pressure inside the sealed chamber; a fumigation unit connected to the sealed chamber via a second pipeline, the fumigation unit including a fumigation gas storage tank and a gas mass flow controller for controlling the fumigation gas flow rate; and an environmental control unit including a refrigeration unit connected to the sealed chamber via a third pipeline, a humidifier connected to the sealed chamber via a fourth pipeline, and a... The system includes a temperature sensor and a humidity sensor located within a sealed chamber; a control unit electrically connected to a vacuum pump, a pressure sensor, a gas mass flow controller, a temperature sensor, a humidity sensor, a refrigeration unit, and a humidifier; the control unit stores a preservation parameter prediction model trained based on a machine learning algorithm; the control unit is configured to receive fruit type information, initial quality parameters, and a target storage period, input these information into the preservation parameter prediction model, and output the target temperature value, target humidity value, target pressure value, and fumigation duration.

[0007] Furthermore, the fumigation unit also includes a gas mixing device and a circulation loop; the gas mixing device's inlet end is connected to different fumigation gas storage tanks through multiple branch pipes, and each branch pipe is equipped with a gas mass flow controller; the gas mixing device's outlet end is connected to the sealed chamber through a second pipe; the circulation loop includes a circulation pipe connecting the top and bottom of the sealed chamber, and a circulation fan installed on the circulation pipe.

[0008] Furthermore, the initial quality parameters include fruit firmness, soluble solids content, color parameters, and initial weight.

[0009] Furthermore, the control unit also includes a preprocessing module and a feature extraction module; the preprocessing module is configured to embed the fruit variety information to generate a variety feature vector, standardize the initial quality parameters, and calculate the relative deterioration degree relative to the variety standard value; the feature extraction module is configured to construct a quality decay feature matrix based on the variety feature vector and the relative deterioration degree, segment and encode the target storage period according to a preset time granularity, and generate periodic time-series features; the quality decay feature matrix and the periodic time-series features are weighted and fused through an attention mechanism to form an enhanced preservation feature representation, which is used as the input of the preservation parameter prediction model.

[0010] Furthermore, the preservation parameter prediction model includes: multiple base learners, constructed using support vector machines, random forests, and gradient boosting trees, respectively, with the input of each base learner being an enhanced preservation feature representation; a dynamic weighting module, configured to calculate the weight coefficients of each base learner based on fruit variety information and the target storage period; wherein, the output of the preservation parameter prediction model is a weighted fusion of the prediction results of each base learner, and the weight coefficients are dynamically adjusted based on the historical prediction accuracy of each base learner under the corresponding combination of fruit variety and storage period; the preservation parameter prediction model is trained using a cross-variety transfer learning strategy, pre-training the base learners using large sample data of source domain fruit varieties, and then fine-tuning them using small sample data of target domain fruit varieties.

[0011] Furthermore, fruit variety information is mapped to a high-dimensional variety encoding vector, and the target storage period is discretized and encoded according to a preset time interval to generate a periodic encoding vector. The variety encoding vector and the periodic encoding vector are concatenated to form the current query vector. The cosine similarity between the current query vector and the feature vectors of all samples in the historical data is calculated, and the top k historical samples with the highest similarity are selected as the nearest neighbor set. Based on the nearest neighbor set, the prediction accuracy of each base learner on the corresponding historical samples is calculated, and the average accuracy of each base learner is calculated. The average accuracy of each base learner is input into the softmax function for normalization to obtain the weight coefficients of each base learner.

[0012] Furthermore, the control unit receives the target temperature, target humidity, target pressure, and fumigation duration output from the preservation parameter prediction model; it activates the pressure reduction unit to adjust the pressure inside the sealed chamber to the target pressure value via a vacuum pump, and performs closed-loop control based on feedback from the pressure sensor throughout the storage period; simultaneously, it activates the environmental control unit to lower the temperature inside the sealed chamber to the target temperature value via a refrigeration unit and adjust the humidity inside the sealed chamber to the target humidity value via a humidifier, and performs independent closed-loop control based on feedback from the temperature and humidity sensors respectively; when the temperature, humidity, and pressure are all stable within the target range, the fumigation unit is activated to introduce a preset flow rate of fumigation gas into the sealed chamber according to the fumigation duration; after the fumigation duration ends, the introduction of fumigation gas is stopped, and the process enters the maintenance phase, during which the environmental control unit and the pressure reduction unit continue to operate to maintain the temperature, humidity, and pressure within the target range.

[0013] According to another aspect of the present invention, a method for preserving fruit by depressurized fumigation is also provided, wherein the fruit is preserved using the aforementioned depressurized fumigation device.

[0014] The present invention has at least the following beneficial effects: This invention utilizes a preservation parameter prediction model integrated into the control unit to precisely output suitable temperature, humidity, pressure, and fumigation duration based on fruit type, initial quality parameters, and target storage period. This allows for personalized customization of preservation parameters, solving the problems of fixed parameters and poor adaptability in traditional preservation technologies. The decompression unit, fumigation unit, and environmental control unit work collaboratively. The decompression environment reduces the fruit's respiration intensity and metabolic rate; fumigation effectively inhibits pathogen growth; and precise temperature and humidity control reduces moisture loss. These three elements create a synergistic preservation effect, comprehensively alleviating quality deterioration issues such as softening, rotting, and wilting. Each unit achieves closed-loop control through sensor feedback, ensuring a stable and controllable storage environment, significantly reducing human error, and avoiding the risk of fumigation gas residue, thus guaranteeing the safety of the fruit for consumption. This invention significantly extends the storage period and shelf life of fruits, effectively maintaining their initial firmness, flavor, and nutritional components, meeting the needs of long-distance transportation and year-round supply, and enhancing product market competitiveness.

[0015] Other advantages, objectives and features of the present invention will become apparent in part from the following description, and in part from those skilled in the art through study and practice of the invention. Attached Figure Description

[0016] Figure 1 This is a logical framework diagram of one embodiment of this application. Detailed Implementation

[0017] The present invention will now be described in further detail so that those skilled in the art can implement it based on the description.

[0018] It should be understood that terms such as "having," "comprising," and "including" used in the embodiments of this application do not exclude the presence or addition of one or more other elements or combinations thereof. All directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of this application are only used to explain the relative positional relationship and movement of components in a specific posture. If the specific posture changes, the directional indication will also change accordingly. When an element is referred to as "fixed to" or "set on" another element, it can be directly on the other element or may have an intervening element present. When an element is referred to as "connected to" another element, it can be directly connected to the other element or indirectly connected to the other element through an intervening element. Descriptions involving "first," "second," etc., in the embodiments of this application are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of indicated technical features. Therefore, a feature defined with "first" or "second" may explicitly or implicitly include at least one of those features.

[0019] It should be noted that the technical solutions of the various embodiments of this application can be combined with each other, but only if they are based on the ability of those skilled in the art to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such combination of technical solutions does not exist and is not within the scope of protection claimed by this application.

[0020] like Figure 1 As shown, embodiments of this application provide a fruit depressurization fumigation preservation device, comprising: a sealed chamber for containing fruit; a depressurization unit connected to the sealed chamber via a first pipeline, the depressurization unit including a vacuum pump and a pressure sensor for detecting the pressure inside the sealed chamber; a fumigation unit connected to the sealed chamber via a second pipeline, the fumigation unit including a fumigation gas storage tank and a gas mass flow controller for controlling the flow rate of the fumigation gas; an environmental control unit including a refrigerator connected to the sealed chamber via a third pipeline, a humidifier connected to the sealed chamber via a fourth pipeline, a temperature sensor disposed within the sealed chamber, and a humidity sensor disposed within the sealed chamber; and a control unit electrically connected to the vacuum pump, pressure sensor, gas mass flow controller, temperature sensor, humidity sensor, refrigerator, and humidifier, the control unit storing a preservation parameter prediction model trained based on a machine learning algorithm; wherein, the control unit is configured to receive fruit type information, initial quality parameters, and target storage period, and input the fruit type information, initial quality parameters, and target storage period into the preservation parameter prediction model, and output target temperature value, target humidity value, target pressure value, and fumigation duration.

[0021] For example, the fruit depressurization fumigation and preservation device includes a sealed chamber, which is a cavity with airtight isolation properties. The chamber can be made of 304 stainless steel or food-grade polypropylene and is used to hold fruits such as Xinjiang white apricots, Kashgar cherry plums, and Hotan jujubes. The fruits can be laid flat on breathable plastic trays and placed inside the chamber, with a 10 cm gap between the trays to ensure gas circulation. The device is equipped with a depressurization unit, which is connected to the sealed chamber via a first pipeline, which can be a corrosion-resistant rubber tube. The depressurization unit includes a vacuum pump and a pressure sensor. The vacuum pump can be a rotary vane vacuum pump, and the pressure sensor can be a capacitive pressure sensor. Its probe end is fixed to the middle of the inner wall of the sealed chamber for real-time detection of the gas pressure inside the chamber, with a detection accuracy of ±0.01 MPa. The device also includes a fumigation unit connected to the sealed chamber via a second pipeline, which can be a polytetrafluoroethylene (PTFE) pipe. The fumigation unit includes a fumigation gas storage tank (potentially a pressure-resistant stainless steel tank) and a gas mass flow controller. The fumigation gas storage tank can be a pressure-resistant stainless steel tank, and the stored fumigation gas can be chlorine dioxide. The gas mass flow controller can be a thermal mass flow controller for precisely controlling the flow rate of the fumigation gas. The environmental control unit is the core component ensuring a stable chamber environment. It includes a refrigerator connected to the sealed chamber via a third pipeline (potentially an insulated copper pipe) and a compressor refrigerator, which lowers the chamber temperature through refrigerant circulation. It also includes a humidifier connected to the sealed chamber via a fourth pipeline (potentially a plastic flexible hose) and an ultrasonic humidifier, which increases the chamber humidity through atomization. The sealed chamber also contains a temperature sensor and a humidity sensor. The temperature sensor can be a platinum resistance temperature sensor, and the humidity sensor can be a capacitive humidity sensor. Both are fixed in the center of the chamber and are used to collect temperature and humidity data in real time. The temperature detection accuracy can reach ±0.1℃, and the humidity detection accuracy can reach ±1%RH. The control unit can be an embedded controller, electrically connected to a vacuum pump, pressure sensor, gas mass flow controller, temperature sensor, humidity sensor, refrigeration unit, and humidifier via wires. Internally, the control unit stores a preservation parameter prediction model trained using a machine learning algorithm, such as a backpropagation (BP) neural network. This model is generated through extensive experimental data on fruit storage in Xinjiang and includes an input layer, a hidden layer, and an output layer. The hidden layer can have up to 32 nodes. The control unit is configured to receive user input information about the fruit type (e.g., "Kashgar cherry plum"), initial quality parameters, and a target storage period (e.g., 45 days). The control unit first converts this information into digital signals recognizable by the model, then inputs it into the preservation parameter prediction model. The model receives the signals through the input layer, performs weight calculations in the hidden layer, and applies a sigmoid activation function. Finally, the output layer outputs the target temperature (e.g., 3℃), target humidity (e.g., 92%RH), target pressure (e.g., 0.04MPa), and fumigation duration (e.g., 75 minutes).

[0022] In existing technologies, the preservation of fruits such as Xinjiang white apricots and Kashgar cherry plums mostly relies on traditional cellar refrigeration or simple fumigation. Cellar temperatures are greatly affected by external factors, fluctuating within ±3℃. Fumigation is achieved by manually pouring chemicals, making it impossible to control gas concentration and failing to adjust conditions based on the fruit's initial quality and target storage period. This embodiment integrates various functional units through a control unit, using a trained model to precisely match preservation parameters according to the specific conditions of Xinjiang stone fruits. Each unit provides real-time data feedback through sensors, ensuring that parameters such as temperature, humidity, and pressure remain stable within the target range. Compared to existing technologies, this embodiment achieves personalized adaptation of preservation parameters, avoiding the limitations of single technologies. It more comprehensively inhibits the vigorous post-harvest metabolic activity and pathogen growth of Xinjiang stone fruits, better maintaining their initial quality, such as sugar content and flavor. This collaborative approach with precise parameter control solves the problems of fixed parameters and poor adaptability in existing technologies.

[0023] In another embodiment, the fumigation unit further includes a gas mixing device and a circulation loop; the inlet of the gas mixing device is connected to different fumigation gas storage tanks through multiple branch pipes, and each branch pipe is equipped with a gas mass flow controller; the outlet of the gas mixing device is connected to the sealed chamber through a second pipe; the circulation loop includes a circulation pipe connecting the top and bottom of the sealed chamber, and a circulation fan installed on the circulation pipe.

[0024] For example, the fumigation unit of the fruit pressure-reducing fumigation preservation device also includes a gas mixing device and a circulation loop. The gas mixing device can be a static mixer, with its inlet end connected to different fumigation gas storage tanks via 2-3 branch pipes. For example, one tank stores chlorine dioxide gas, and another stores ozone gas. Each branch pipe can be a polytetrafluoroethylene pipe with an inner diameter of 8mm, and each pipe is equipped with a gas mass flow controller, which can be a thermal mass flow controller capable of adjusting the inflow rate of each gas. The outlet end of the gas mixing device is connected to a sealed chamber via a second pipe, through which the mixed gas enters the chamber, achieving the synergistic effect of multiple fumigation gases. The circulation loop includes a circulation pipe connecting the top and bottom of the sealed chamber. The pipe can be a stainless steel pipe with a diameter of 50mm. A circulation fan is installed on the pipe. This fan can be an axial flow fan. After starting, it can cause the gas in the chamber to flow out from the top and return to the bottom through the circulation pipe, forming an internal gas circulation. This makes the fumigation gas and temperature and humidity more evenly distributed in the chamber, which is especially suitable for ensuring the consistency of the treatment of stacked fruits when storing Xinjiang white apricots.

[0025] In existing technologies, fumigation preservation of stone fruits such as Kashgar cherry plums from Xinjiang often employs a single gas, with the gas only entering from one side of the chamber. This can easily lead to excessively high concentrations in some areas while insufficient concentrations are present on the other, resulting in significant variations in fruit preservation effectiveness. This embodiment utilizes a gas mixing device to proportionally distribute multiple gases, while a circulation loop promotes uniform gas diffusion. This ensures that all Xinjiang stone fruits within the chamber are exposed to a suitable concentration of fumigant gas, preventing localized over- or under-treatment. Compared to existing technologies, this approach more comprehensively inhibits pathogens and maintains more consistent fruit quality, resolving the problems of uneven gas distribution and the limitations of single-gas effects. In another embodiment, the initial quality parameters include fruit firmness, soluble solids content, color parameters, and initial weight. For example, in this fruit decompression fumigation preservation device, the control unit receives the initial quality parameters including fruit firmness, soluble solids content, color parameters, and initial weight. Fruit firmness can be measured using a fruit firmness tester, where the probe is vertically inserted into the center of the flesh of a Xinjiang Hotan jujube, avoiding the pit. The unit is N / cm², and the measured value could be 7.2 N / cm². Soluble solids content can be determined using a handheld refractometer, by dropping fruit juice onto a detection prism, and the reading is in percentage (%). For example, the soluble solids content of a Xinjiang white apricot is measured to be 14.5%. Color parameters can be detected using a colorimeter, recording the L value (brightness), a value (red-green hue), and b value (yellow-blue hue). Initial weight can be measured using an electronic balance, accurate to 0.1g; for example, the initial weight of a single Xinjiang white apricot is 45.6g. These parameters collectively reflect the freshness and quality of the fruit after harvest, providing basic data for the preservation parameter prediction model.

[0026] This embodiment, by collecting multi-dimensional initial quality parameters, can more comprehensively reflect the physiological state of the fruit, making the input of the preservation parameter prediction model more accurate. Compared with existing technologies, it can adjust the preservation strategy in a targeted manner, avoiding over- or under-preservation due to incomplete parameters, and better maintaining the original quality of Xinjiang stone fruits.

[0027] In another embodiment, the control unit further includes a preprocessing module and a feature extraction module; the preprocessing module is configured to embed the fruit variety information into a representation, generate a variety feature vector, standardize the initial quality parameters, and calculate the relative deterioration degree relative to the variety standard value; the feature extraction module is configured to construct a quality decay feature matrix based on the variety feature vector and the relative deterioration degree, segment and encode the target storage period according to a preset time granularity, and generate periodic time-series features; and perform weighted fusion of the quality decay feature matrix and the periodic time-series features through an attention mechanism to form an enhanced preservation feature representation, which is used as the input to the preservation parameter prediction model.

[0028] For example, the control unit of this fruit vacuum fumigation preservation device also includes a preprocessing module and a feature extraction module. The preprocessing module is configured to embed fruit variety information, such as converting the variety information of "Xinjiang White Apricot" into a 16-dimensional variety feature vector, where each element corresponds to a specific attribute of the variety; simultaneously, it standardizes the initial quality parameters using the z-score standardization formula (standardized value = (measured value - mean) / standard deviation), for example, taking the measured fruit firmness as 7.2 N / cm. 2 The hardness is converted to a standardized value of 0.8, and then the relative deterioration degree relative to the standard value of the variety is calculated using the formula (standard value of variety - measured value) / standard value of variety. If the standard hardness of Xinjiang white apricot is 8.0 N / cm², then the relative deterioration degree is (8.0 - 7.2) / 8.0 = 0.1. The feature extraction module is configured to construct a 3x5 matrix of quality degradation features (3 parameters, 5 time features) based on the variety feature vector and the relative deterioration degree. The target storage period of 45 days is divided into 9 segments with a time granularity of 5 days, and a 9-dimensional periodic time feature is generated using one-hot encoding. The weight coefficients of the quality degradation feature matrix and the periodic time feature are calculated through an attention mechanism. For example, a weight of 0.3 is assigned to the hardness feature, and a weight of 0.2 is assigned to the third segment of the period. After weighted fusion, an enhanced preservation feature representation is formed, which serves as the input to the preservation parameter prediction model.

[0029] In existing technologies, preservation parameters are often calculated using raw data directly without feature transformation of variety information and quality parameters, making it difficult for the model to capture key patterns. This embodiment transforms abstract information into computable features through preprocessing and feature extraction. The attention mechanism further highlights important features, making the model input more representative. Compared with existing technologies, this improves the accuracy of parameter prediction, and is especially suitable for the diverse varieties of Xinjiang fruits. In another embodiment, the preservation parameter prediction model includes: multiple base learners, constructed using support vector machines, random forests, and gradient boosting trees, respectively, with the input of each base learner being an enhanced preservation feature representation; a dynamic weighting module configured to calculate the weight coefficients of each base learner based on fruit variety information and the target storage period; wherein, the output of the preservation parameter prediction model is a weighted fusion of the prediction results of each base learner, and the weight coefficients are dynamically adjusted based on the historical prediction accuracy of each base learner under the corresponding combination of fruit variety and storage period; the preservation parameter prediction model is trained using a cross-variety transfer learning strategy, pre-training the base learners using large sample data of source domain fruit varieties, and then fine-tuning them using small sample data of target domain fruit varieties.

[0030] For example, the preservation parameter prediction model of this fruit decompression fumigation preservation device includes multiple base learners and a dynamic weighting module. The multiple base learners are constructed using Support Vector Machine (SVM), Random Forest, and Gradient Boosting Tree (GPRS) algorithms, respectively. The SVM uses the RBF kernel function with a penalty coefficient of 10; the Random Forest contains 50 decision trees with a maximum depth of 8; and the GPRS has a learning rate of 0.1 and 100 iterations. The input to each base learner is an enhanced preservation feature representation. The dynamic weighting module is configured to calculate the weight coefficients of each base learner based on the fruit type information and the target storage period. For example, for Xinjiang white apricots and a 45-day storage period, the calculated weights are: SVM 0.3, Random Forest 0.4, and GPRS 0.3. The output of the preservation parameter prediction model is a weighted fusion of the prediction results from each base learner. For example, if the support vector machine predicts a temperature of 3.2℃, the random forest predicts 3.0℃, and the gradient boosting tree predicts 2.8℃, the fused target temperature is 3.2×0.3+3.0×0.4+2.8×0.3=3.0℃. The weight coefficients are dynamically adjusted based on the historical prediction accuracy of each base learner for the corresponding variety and cycle. If the historical accuracy of the gradient boosting tree in the 45-day cycle of Xiaobai apricot increases from 82% to 85%, its weight will be increased accordingly. The model is trained through cross-variety transfer learning. First, the base learners are pre-trained using 1000 sets of samples from the source domain (e.g., common peach), and then fine-tuned using 200 sets of small samples from the target domain (e.g., Xinjiang Xiaobai apricot) to improve the model's adaptability to niche varieties. In existing technologies, the prediction of stone fruit preservation parameters often employs a single model, trained only on data from a single variety, resulting in poor adaptability to small sample data of stone fruits unique to Xinjiang. This embodiment, through multi-base learner fusion and dynamic weighting, combined with cross-variety transfer learning, integrates the advantages of different algorithms and utilizes data from similar varieties to improve prediction performance under small sample conditions. Compared to existing technologies, it is more suitable for predicting the preservation parameters of diverse stone fruits in Xinjiang.

[0031] In another embodiment, fruit type information is mapped to a high-dimensional variety encoding vector, and the target storage period is discretized and encoded according to a preset time interval to generate a period encoding vector; the variety encoding vector and the period encoding vector are concatenated to form the current query vector; the cosine similarity between the current query vector and the feature vectors of all samples in the historical data is calculated, and the top k historical samples with the highest similarity are selected as the nearest neighbor set; based on the nearest neighbor set, the prediction accuracy of each base learner on the corresponding historical samples is calculated, and the average accuracy of each base learner is calculated; the average accuracy of each base learner is input into the softmax function for normalization processing to obtain the weight coefficients of each base learner.

[0032] For example, in this fruit decompression fumigation preservation device, when the control unit calculates the weight coefficients of the base learners, it first maps the fruit type information into a high-dimensional variety encoding vector. For example, "Kashgar cherry plum" is mapped into a 128-dimensional numerical vector, with each dimension corresponding to the genetic or physiological characteristics of the variety. The target storage period of 45 days is discretized into three intervals with a preset time interval of 15 days, generating a 3-dimensional periodic encoding vector, such as [0,1,0] representing the interval between 15 and 30 days. The variety encoding vector and the periodic encoding vector are concatenated to form a 131-dimensional current query vector. The cosine similarity between this vector and the feature vectors of all samples in the historical data is calculated using the formula cosθ=(A·B) / (|A|·|B|), where A is the current query vector and B is the historical sample vector. The top 5 historical samples with the highest similarity are selected as the nearest neighbor set. Based on this set, the prediction accuracy of each base learner on these samples is calculated. For example, the average accuracy of the support vector machine on the 5 samples is 84%, that of the random forest is 86%, and that of the gradient boosting tree is 88%. These average accuracies are normalized by inputting them into the softmax function, i.e., weight coefficient = exp(accuracy) / sum(exp(each accuracy)), and the weights of the support vector machine, random forest, and gradient boosting tree are calculated to be 0.31, 0.32, and 0.37 respectively. This embodiment selects similar historical samples using cosine similarity and dynamically calculates weights to make the weights more consistent with the current characteristics of the fruit and its cycle. Compared with existing technologies, this can improve the adaptability of the prediction results, and is especially effective for Xinjiang stone fruits with large differences in storage cycles. In another embodiment, the control unit receives the target temperature, target humidity, target pressure, and fumigation duration output from the preservation parameter prediction model; it activates the decompression unit to adjust the pressure inside the sealed chamber to the target pressure value via a vacuum pump, and performs closed-loop control based on feedback from the pressure sensor throughout the storage period; simultaneously, it activates the environmental control unit to lower the temperature inside the sealed chamber to the target temperature value via a refrigeration unit and adjust the humidity inside the sealed chamber to the target humidity value via a humidifier, and performs independent closed-loop control based on feedback from the temperature and humidity sensors respectively; when the temperature, humidity, and pressure are all stable within the target range, it activates the fumigation unit to introduce a preset flow rate of fumigation gas into the sealed chamber according to the fumigation duration; after the fumigation duration ends, it stops introducing fumigation gas and enters the maintenance phase, during which the environmental control unit and the decompression unit continue to operate to maintain the temperature, humidity, and pressure within the target range.

[0033] For example, the control unit of the fruit depressurization fumigation preservation device receives the target temperature value of 3℃, the target humidity value of 92%RH, the target pressure value of 0.04MPa, and the fumigation duration of 75 minutes from the preservation parameter prediction model. The depressurization unit is activated, controlling the vacuum pump to extract air from the sealed chamber through the first pipeline. The pressure sensor monitors the pressure in real time. Extraction stops when the pressure is below 0.04MPa and restarts when it is above 0.045MPa. This closed-loop control maintains stable pressure throughout the storage period. Simultaneously, the environmental control unit is activated. The refrigeration unit delivers cooling energy through the third pipeline. Cooling is increased when the temperature sensor detects a chamber temperature above 3.2℃ and decreased when it is below 2.8℃. The humidifier releases water mist through the fourth pipeline. Humidification is activated when the humidity sensor detects a humidity below 90%RH and stops when it is above 94%RH. Both units maintain stable temperature and humidity through independent closed-loop control. After the temperature, humidity, and pressure have stabilized within ±5% of the target values ​​for 3 minutes, the fumigation unit is started. The gas mass flow controller introduces fumigation gas into the chamber through the second pipeline at a preset flow rate for 75 minutes. After fumigation, the introduction of fumigation gas is stopped, and the process enters the maintenance phase. The environmental control unit and pressure reduction unit continue to operate, collecting temperature, humidity, and pressure data every 10 minutes to ensure that the parameters are maintained within the target range until the end of the storage period.

[0034] In existing technologies, the control of temperature, humidity, pressure, and fumigation for preservation is mostly manual, with asynchronous adjustments of various parameters, often resulting in excessive temperature fluctuations during fumigation. This embodiment achieves coordinated stability of various parameters through automated closed-loop control. Fumigation is initiated only after the environmental conditions meet the standards, and continuous monitoring is performed during the maintenance phase. Compared with existing technologies, this avoids human error, ensures a stable preservation environment, and better extends the storage time of Xinjiang stone fruits.

[0035] The following is a description of a specific embodiment.

[0036] The experimental group used Xinjiang white apricots as the preservation target. The specific preservation method was as follows: Xinjiang white apricots harvested within 24 hours were first sorted and graded, removing damaged or diseased fruits. They were then laid flat on breathable plastic trays (leaving a 10 cm gap between trays) and placed in a 3 cubic meter sealed 304 stainless steel chamber. The fruit type "Xinjiang white apricot," initial quality parameters (fruit firmness 7.0 N / cm², soluble solids content 14.8%, color parameters L=33.5, a=24.2, b=18.6, initial single fruit weight 43.2 g) and the target storage period of 45 days were input into the control unit. The preprocessing module of the control unit embeds the variety information into a 16-dimensional variety feature vector. After z-score standardization of the initial quality parameters, it calculates the relative deterioration degree (hardness relative deterioration degree 0.08). The feature extraction module constructs a quality decay feature matrix based on the above data, and encodes the 45-day storage period into time-series features in 5-day granularity segments. These features are then weighted and fused using an attention mechanism to form enhanced preservation features. After receiving these features, the three base learners (support vector machine, random forest, and gradient boosting tree) of the preservation parameter prediction model receive the features. The dynamic weighting module selects five nearest-neighbor historical samples using cosine similarity, calculates the weights of each base learner (0.3, 0.4, 0.3), and outputs the target temperature of 3℃, target humidity of 92%RH, target pressure of 0.04MPa, and fumigation duration of 75 minutes. The pressure reduction unit is activated, and the chamber pressure is adjusted to the target value by a rotary vane vacuum pump. The pressure sensor provides real-time feedback to achieve closed-loop control. Simultaneously, the environmental control unit is activated, and the compressor refrigerator and ultrasonic humidifier maintain stable temperature and humidity through independent closed-loop control. After the three parameters stabilize for 3 minutes, the fumigation unit is activated. The gas mixing device mixes chlorine dioxide and ozone in a ratio of 2:1. The axial flow fan in the circulation loop promotes uniform gas diffusion. After continuous fumigation for 75 minutes, the system enters the maintenance phase. All units continue to operate until the storage is completed.

[0037] The first control group used a single cold storage preservation method: Xinjiang white apricots of the same batch and initial quality as the experimental group were selected, screened and graded, and placed in a regular cold storage. The temperature was set at 3°C ​​and the humidity at 92%RH, without pressure reduction treatment or fumigation treatment. During the storage period, the temperature and humidity were recorded manually every day without dynamic control, and the storage period was 45 days.

[0038] The second control group used the traditional fumigation + cold storage preservation method: Xinjiang white apricots of the same batch were selected, screened and graded and placed in a regular cold storage (temperature 3℃, humidity 92%RH). On the first day of storage, chlorine dioxide was manually poured for fumigation (without flow control) and no pressure reduction treatment was performed. During the storage period, only the temperature and humidity were manually adjusted periodically. The parameters were not adjusted according to the initial quality of the fruit and the storage period. The storage period was 45 days.

[0039] The blank group did not use any preservation measures: Xinjiang white apricots from the same batch were selected, screened and graded, and then placed directly in a room temperature environment (temperature around 25℃, humidity fluctuating naturally) without sealing, pressure reduction, fumigation, or temperature and humidity control, and left for 45 days.

[0040] After 45 days of storage, two core parameters were measured in the four samples: fruit firmness and soluble solids content. The fruit firmness of the experimental group was 5.8 N / cm. 2 The soluble solids content was 13.6%; the fruit firmness of the first control group was 4.2 N / cm. 2 The soluble solids content was 11.5%; the fruit firmness of the second control group was 4.5 N / cm. 2 The soluble solids content was 12.1%; the fruit firmness of the control group was 2.1 N / cm. 2 The soluble solids content was 9.8%. It can be seen that the decline in both parameters in the experimental group was significantly less than that in the control group and the blank group. The first control group, lacking the synergistic effect of decompression and fumigation, struggled to inhibit fruit metabolism and pathogen growth, resulting in a rapid decrease in firmness. The second control group, although fumigated, lacked precise flow control and a decompression environment, and did not incorporate parameters tailored to the initial fruit quality, thus limiting its preservation effect. The blank group exhibited the most severe natural deterioration. The experimental group, by integrating decompression, precise fumigation, and synergistic temperature and humidity control, and utilizing a machine learning-based parameter prediction model for personalized adaptation, while optimizing the model output through feature extraction and dynamic weighting, solved the limitations of existing technologies such as single-method limitations, fixed parameters, and poor adaptability. Its combination of multi-unit synergy and precise parameter control resulted in superior preservation effects.

[0041] Although embodiments of the present invention have been disclosed above, they are not limited to the applications listed in the specification and embodiments. They can be applied to various fields suitable for the present invention. For those skilled in the art, other modifications can be easily made. Therefore, without departing from the general concept defined by the claims and their equivalents, the present invention is not limited to the specific details and embodiments shown and described herein.

Claims

1. A fruit decompression fumigation and preservation device, characterized in that, include: A sealed chamber for holding fruit; The pressure reduction unit is connected to the sealed chamber via a first pipeline. The pressure reduction unit includes a vacuum pump and a pressure sensor for detecting the pressure inside the sealed chamber. The fumigation unit is connected to the sealed chamber via a second pipeline. The fumigation unit includes a fumigation gas storage tank and a gas mass flow controller for controlling the flow rate of the fumigation gas. The environmental control unit includes a refrigeration unit connected to a sealed chamber via a third pipeline, a humidifier connected to the sealed chamber via a fourth pipeline, a temperature sensor installed in the sealed chamber, and a humidity sensor installed in the sealed chamber. The control unit is electrically connected to the vacuum pump, pressure sensor, gas mass flow controller, temperature sensor, humidity sensor, refrigerator and humidifier. The control unit stores a preservation parameter prediction model trained based on machine learning algorithm. The control unit is configured to receive fruit type information, initial quality parameters, and target storage period, and input the fruit type information, initial quality parameters, and target storage period into the preservation parameter prediction model, and output the target temperature value, target humidity value, target pressure value, and fumigation duration.

2. The fruit decompression fumigation and preservation device as described in claim 1, characterized in that, The fumigation unit also includes a gas mixing device and a circulation loop; The gas mixing device’s inlet is connected to different fumigation gas storage tanks via multiple branch pipes. Each branch pipe is equipped with a gas mass flow controller. The gas mixing device’s outlet is connected to a sealed chamber via a second pipe. The circulation loop includes a circulation pipe that connects the top and bottom of the sealed chamber, and a circulation fan installed on the circulation pipe.

3. The fruit decompression fumigation and preservation device as described in claim 1, characterized in that, Initial quality parameters include fruit firmness, soluble solids content, color parameters, and initial weight.

4. The fruit decompression fumigation and preservation device as described in claim 1, characterized in that, The control unit also includes a preprocessing module and a feature extraction module; The preprocessing module is configured to embed the fruit variety information, generate a variety feature vector, standardize the initial quality parameters, and calculate the relative degree of deterioration relative to the standard value of the variety. The feature extraction module is configured to construct a quality decay feature matrix based on the variety feature vector and relative deterioration degree, and to segment and encode the target storage period according to a preset time granularity to generate periodic time features. The quality decay feature matrix and periodic time-series features are weighted and fused using an attention mechanism to form an enhanced preservation feature representation, which is then used as input to the preservation parameter prediction model.

5. The fruit decompression fumigation and preservation device as described in claim 1, characterized in that, Preservation parameter prediction models include: Multiple base learners are constructed using support vector machines, random forests, and gradient boosting trees, respectively. The input to each base learner is an enhanced fresh feature representation. The dynamic weighting module is configured to calculate the weight coefficients of each base learner based on fruit type information and target storage period. The output of the preservation parameter prediction model is a weighted fusion of the prediction results of each base learner. The weight coefficients are dynamically adjusted by the historical prediction accuracy of each base learner under the corresponding combination of fruit varieties and storage cycles. The preservation parameter prediction model is trained by a cross-variety transfer learning strategy. The base learners are pre-trained using large sample data of source domain fruit varieties and then fine-tuned using small sample data of target domain fruit varieties.

6. The fruit decompression fumigation and preservation device as described in claim 1, characterized in that, Fruit type information is mapped into a high-dimensional variety coding vector, and the target storage period is discretized and coded according to a preset time interval to generate a period coding vector; The variety code vector and the period code vector are combined to form the current query vector; Calculate the cosine similarity between the current query vector and the feature vectors of all samples in the historical data, and select the k historical samples with the highest similarity as the nearest neighbor set; Based on the nearest neighbor set, the prediction accuracy of each base learner on the corresponding historical samples is statistically analyzed, and the average accuracy of each base learner is calculated. The average accuracy of each base learner is input into the softmax function for normalization to obtain the weight coefficients of each base learner.

7. The fruit decompression fumigation and preservation device as described in claim 1, characterized in that, The control unit receives the target temperature, target humidity, target pressure, and fumigation duration output by the preservation parameter prediction model. The pressure reduction unit is activated, and the pressure in the sealed chamber is adjusted to the target pressure value through a vacuum pump. Closed-loop control is performed through feedback from the pressure sensor throughout the storage period. The environmental control unit is activated simultaneously. The refrigeration unit lowers the temperature inside the sealed chamber to the target temperature value, and the humidifier adjusts the humidity inside the sealed chamber to the target humidity value. Independent closed-loop control is performed based on feedback from the temperature sensor and humidity sensor, respectively. When the temperature, humidity and pressure are stable within the target range, start the fumigation unit and introduce a preset flow rate of fumigation gas into the sealed chamber according to the fumigation maintenance time; After the fumigation maintenance time ends, the fumigation gas supply is stopped, and the process enters the maintenance phase. During this phase, the environmental control unit and the pressure reduction unit continue to operate to maintain the temperature, humidity, and pressure within the target range.

8. A method for preserving fruit using decompression fumigation, characterized in that, Fruit preservation is performed using the fruit decompression fumigation preservation device according to any one of claims 1-7.

Images