Spectrum optimization system for postharvest preservation of leaf vegetables

Abstract

The invention relates to the technical field of spectrum optimization and intelligent illumination, in particular to a spectrum optimization system for fresh keeping of picked leaf vegetables, comprising a light source module which is composed of a plurality of independent and controllable LED units with different wavelengths and is used for generating a composite spectrum for irradiating the leaf vegetables; the spectrum sensing module is directly aligned with the light emitting surface of the light source module and is used for collecting the composite spectrum in real time and outputting multi-channel spectrum data; the control module is connected with the spectrum sensing module and the light source module; the optimal preservation spectrum database is integrated in a memory of the control module and stores target spectrum data corresponding to different leaf vegetable types; according to the invention, a closed-loop control system composed of a spectrum sensing module, a spectrum reconstruction unit, a spectrum matching unit and a spectrum optimization unit is constructed, so that the technical span from fixed spectrum illumination to dynamic precise spectrum regulation and control is realized, and the technical problem that the traditional fixed spectrum cannot adapt to the physiological characteristics of different leaf vegetables is effectively solved.

Description

Technical Field

[0001] This invention relates to the fields of spectral optimization and intelligent lighting technology, specifically a spectral optimization system for post-harvest preservation of leafy vegetables. Background Technology

[0002] Post-harvest preservation of leafy green vegetables is a significant challenge for the food logistics and retail industry. Vegetables remain living organisms after harvesting, continuing respiration and transpiration, leading to chlorophyll degradation, tissue wilting, and nutrient loss. This ultimately results in yellowing and spoilage, causing substantial economic losses and food waste.

[0003] Currently, the mainstream vegetable preservation technologies can be mainly divided into chemical preservation methods and physical preservation methods. Chemical preservation methods carry the risk of chemical reagent residues, which can easily cause consumers to worry about food safety. Physical preservation methods, such as low-temperature refrigeration, have high energy consumption, and inappropriate low temperatures can easily lead to chilling injury. Modified atmosphere packaging has high requirements for packaging materials and equipment, and is therefore more expensive.

[0004] In recent years, light environment control, as an emerging physical preservation technology, has shown great application potential. Light-emitting diode (LED) light sources, due to their advantages such as wavelength selectivity, high luminous efficiency, cold light source, and ease of control, have been widely used in this field. However, existing LED preservation technologies still have significant drawbacks: Existing technologies often use pre-set LED combinations with fixed wavelengths and light intensities (e.g., a fixed ratio of red to blue light) for irradiation; however, different types of leafy green vegetables have different chlorophyll contents and ratios, and a fixed spectrum cannot adapt to these variables, resulting in unstable preservation effects and difficulty in achieving optimal results.

[0005] Most current systems are open-loop systems, lacking a monitoring and feedback mechanism for the actual output spectrum. They cannot detect and correct spectral drift caused by LED driving current, ambient temperature, and device aging, resulting in the actual irradiation spectrum gradually deviating from the originally designed optimal spectrum, and the preservation effect decreasing accordingly.

[0006] Therefore, there is an urgent need in this field for a spectral optimization preservation system that can achieve dynamic and accurate spectral matching, has multi-species adaptive capabilities, and is stable over the long term. Summary of the Invention

[0007] The purpose of this invention is to overcome the shortcomings of the prior art and provide a spectral optimization system and method for postharvest preservation of leafy vegetables. This system can dynamically output and maintain the optimal preservation spectrum that matches different leafy vegetable varieties in real time, thereby achieving a precise, efficient and stable postharvest preservation effect.

[0008] To achieve the above objectives, the present invention provides the following technical solution: a spectral optimization system for post-harvest preservation of leafy vegetables, comprising: The light source module consists of multiple independently controllable LED units of different wavelengths, used to generate a composite spectrum for irradiating leafy vegetables; The spectral sensing module is directly aligned with the light-emitting surface of the light source module, and is used to acquire the composite spectrum in real time and output multi-channel spectral data. The control module is connected to the spectral sensing module via an I²C bus and to the light source module via multiple PWM signal lines; An optimal preservation spectral database is integrated into the memory of the control module, storing target spectral data corresponding to different leafy vegetable varieties; The control module includes interconnected functional units: The spectral reconstruction unit is used to receive multi-channel spectral data output by the spectral sensing module and generate high-resolution real-time spectra through the built-in spectral reconstruction model. A spectral matching unit is used to receive the high-resolution real-time spectrum output by the spectral reconstruction unit and perform a matching degree analysis with the target spectral data retrieved from the optimal preservation spectral database. The spectral optimization unit is used to calculate the optimal driving parameters of each LED unit through an optimization algorithm when the matching degree output by the spectral matching unit is lower than a preset threshold, and output the driving parameters to the light source module. The spectral sensing module, the spectral reconstruction unit, the spectral matching unit, the spectral optimization unit, and the light source module in the control module are connected in sequence to form a closed-loop control system, which realizes real-time monitoring and dynamic adjustment, so that the composite spectrum output by the light source module matches the target spectral data.

[0009] Preferably, the spectral sensing module is a spectral sensor with eight visible light channels, and the multi-channel spectral data is the light intensity values ​​of the eight channels output by the spectral sensor. The light source module contains multiple independently controllable LED units with different wavelengths, including four monochromatic LEDs with center wavelengths of 400nm±5nm, 430nm±5nm, 451nm±5nm, and 645nm±5nm, and at least one white LED for providing a continuous background spectrum. Each LED unit is connected to the control module through an independent constant current drive circuit.

[0010] Preferably, in order to achieve a uniform spectral distribution, the LEDs on the substrate of the light source module are arranged in a concentric circle staggered arrangement. Specifically, the LEDs of different wavelengths are arranged on multiple concentric rings, and the LEDs of different wavelengths are arranged alternately on any ring and between adjacent rings, and the center distance between each LED is 20mm to 30mm.

[0011] Preferably, the spectral reconstruction model built into the spectral reconstruction unit is a neural network model. This model takes the spectral data of the 8 channels output by the spectral sensing module as input and maps it to generate high-resolution real-time spectral data with 160 data points. The 160 data points are arranged in wavelength order, uniformly covering the visible light band from 380nm to 780nm, and each data point represents its normalized light intensity value at the corresponding wavelength.

[0012] Preferably, the neural network model adopts a multilayer perceptron structure, with an input layer of 8 nodes, each corresponding to one of the 8 channels of the spectral sensing module; and an output layer of 160 nodes, the output values ​​of which sequentially constitute a high-resolution real-time spectral sequence covering the 380nm to 780nm band and having a spectral resolution of 2.5nm.

[0013] Preferably, the target spectral data in the optimal preservation spectral database is obtained through the following method: The optimal preservation spectrum for each leafy vegetable variety was determined based on light experiments and post-harvest quality analysis of different leafy vegetable samples. The illumination experiment includes applying different spectral energy distributions to sample groups and monitoring multiple post-harvest quality indicators at multiple time points. By conducting statistical analysis and multi-objective optimization on the multiple post-harvest quality indicators, the spectral energy distribution curve that achieves the best overall preservation effect is determined. The spectral energy distribution curve forms energy peaks in the blue and red light regions corresponding to the chlorophyll absorption peaks, and suppresses the energy ratio of the far-red light band. The spectral energy distribution curve is discretized into a predetermined number of data points in the wavelength range of 380-780nm to form target spectral data for each leafy vegetable species. The postharvest quality indicators include chlorophyll content, weight loss rate, and sensory quality indicators.

[0014] Preferably, the neural network model is constructed through the following process: Training data is collected based on multiple mixed spectral samples generated by the light source module. Different spectra are generated by adjusting the PWM weight combination of each LED unit, and the spectral data of the 8 channels of the spectral sensing module and the high-resolution true spectral data as the training target are collected simultaneously. The multilayer perceptron structure is used for model training, with the collected spectral data from 8 channels as input and the corresponding high-resolution spectral ground truth data as the training target. The trained neural network model is output. After format conversion and quantization, the neural network model is integrated into the embedded system of the control module and run to reconstruct the spectral data of the 8 channels of the spectral sensing module into a 160-point high-resolution spectrum in real time.

[0015] Preferably, the spectral matching unit is configured to use a coefficient of determination R² as a quantitative index of the matching degree, wherein R² is calculated using the following formula: ; in, This represents the spectral intensity value of the target spectrum at the i-th wavelength. This represents the spectral intensity value at the i-th wavelength point in the reconstructed real-time spectrum. This represents the average intensity of the target spectrum across all wavelengths, where i is the wavelength index, ranging from 1 to 160. The preset threshold is set to 98%. When R² ≥ 98%, the spectrum is considered to be well matched. When R² < 98%, the spectrum optimization unit is triggered. The spectral optimization unit is configured to use the least squares method as the optimization algorithm, and its objective function is: minΣ[ (λ) - (λ)]²; Where λ is the wavelength variable. (λ) represents the target spectral intensity at wavelength λ. (λ) represents the real-time spectral intensity at wavelength λ; The optimization variables are the weight coefficients W1, W2, W3, W4, and W5 of the 5 LED units, where each weight coefficient Wi corresponds to the luminous intensity weight of one LED. The constraint condition is that for all i, 0 ≤ Wi ≤ 1, and the weighting coefficient Wi is proportional to the PWM duty cycle driving the LED. The spectral optimization unit is further configured to convert the optimal weight coefficient set {W1, W2, W3, W4, W5} obtained by optimization calculation into corresponding PWM duty cycle signals and output them to each LED driving circuit of the light source module.

[0016] Preferably, the workflow of the closed-loop control system is as follows: after the spectral optimization unit outputs new driving parameters to the light source module, the control module starts a 100-500ms delay program to wait for the output spectrum to reach a stable state. After the delay ends, the control module controls the spectral sensing module to perform a new round of spectral data acquisition; The collected spectral data are sequentially transmitted to the spectral reconstruction unit for high-resolution spectral reconstruction, and then transmitted to the spectral matching unit for matching degree verification with the target spectrum; When the verification result shows that the matching degree has not reached the preset threshold, the control module calls the spectral optimization unit again to perform iterative calculation and generate updated driving parameters; This process is repeated until the matching degree output by the spectral matching unit reaches or exceeds a preset threshold, forming a complete closed-loop feedback optimization mechanism.

[0017] Preferably, it also includes a user interface module, the user interface module comprising: The selection unit includes a selection button for cyclically selecting vegetable types and a confirmation button for confirming the selection. The display unit includes a digital tube display for displaying the number of the currently selected vegetable type; The status indicator unit includes a dual-color LED indicator for indicating the system's operating status; The control module responds to the instruction input by the user through the selection unit by retrieving the target spectral data corresponding to the selected vegetable type number from the optimal preservation spectral database. The system firmware pre-stores optimal preservation spectral data for at least 10 kinds of leafy green vegetables, with each vegetable corresponding to a unique number. The control module calls up the corresponding target spectral data according to the number selected by the user.

[0018] Compared with the prior art, the beneficial effects of the present invention are: 1. This invention achieves a technological leap from fixed-spectrum illumination to dynamic and precise spectral control by constructing a closed-loop control system composed of a spectral sensing module, a spectral reconstruction unit, a spectral matching unit, and a spectral optimization unit. It effectively solves the technical problem that traditional fixed-spectrum systems cannot adapt to the physiological characteristics of different leafy vegetable varieties. The system integrates a database of optimal preservation spectral data for various leafy vegetables and a user interface module, and has multi-species adaptive capabilities. Users can select different vegetable varieties through a simple interface, and the system automatically calls the corresponding target spectral formula, significantly improving the versatility and practicality of the equipment.

[0019] 2. This invention adopts a real-time spectral sensing and feedback control mechanism. The system can continuously monitor the actual output spectrum and perform matching degree analysis with the target spectrum, automatically correcting spectral drift caused by LED light source driving current, ambient temperature and device aging, ensuring spectral stability under long-term operation. Through a spectral reconstruction model based on neural networks, the system uses an 8-channel low-cost sensor in conjunction with a lightweight neural network to achieve accurate reconstruction from low-dimensional sensor data to high-resolution spectral data, significantly reducing system hardware costs while ensuring performance.

[0020] 3. This invention utilizes an optimized design of concentrically arranged LED light source modules. LEDs of different wavelengths are alternately arranged on multiple concentric rings, effectively promoting spatial mixing of photons across different wavelengths, avoiding local spectral deviations, and ensuring a high degree of consistency in the spectral distribution on the target irradiation surface. This provides a uniform and stable lighting environment for post-harvest preservation of leafy vegetables. The synergistic effect of these technical features significantly improves the system's preservation effect, operational stability, and economic efficiency. Attached Figure Description

[0021] Figure 1 This is a schematic diagram of the overall structure of the system of the present invention; Figure 2 This is a schematic diagram of the hardware connection of the control module in this invention; Figure 3 This is a schematic diagram of the working process of the closed-loop control system in this invention; Figure 4 This is a schematic diagram of the LED arrangement of the light source module of the present invention; Figure 5 This is a comparison diagram showing the effects of the present invention. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. 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.

[0023] Please see Figures 1-5 The present invention provides an embodiment of a spectral optimization system for post-harvest preservation of leafy vegetables; like Figure 1 As shown, this system adopts a modular design, including a light source module, a spectral sensing module, a control module, an optimal preservation spectrum database, and a user interface module. Each module forms a complete closed-loop control system through clearly defined physical connections and logical relationships. Specifically: The light source module consists of multiple independently controllable LED units of different wavelengths, used to generate a composite spectrum for irradiating leafy vegetables; The spectral sensing module is directly aligned with the light-emitting surface of the light source module, and is used to acquire the composite spectrum in real time and output multi-channel spectral data. The control module is connected to the spectral sensing module via an I²C bus for real-time transmission of 8-channel spectral data and is connected to the light source module via 5 PWM signal lines. An optimal preservation spectral database is integrated into the memory of the control module, storing target spectral data corresponding to different leafy vegetable varieties; The control module includes interconnected functional units: The spectral reconstruction unit is used to receive multi-channel spectral data output by the spectral sensing module and generate high-resolution real-time spectra through the built-in spectral reconstruction model. A spectral matching unit is used to receive the high-resolution real-time spectrum output by the spectral reconstruction unit and perform a matching degree analysis with the target spectral data retrieved from the optimal preservation spectral database. The spectral optimization unit is used to calculate the optimal driving parameters of each LED unit through an optimization algorithm when the matching degree output by the spectral matching unit is lower than a preset threshold, and output the driving parameters to the light source module. The spectral sensing module, the spectral reconstruction unit, the spectral matching unit, the spectral optimization unit, and the light source module in the control module are connected in sequence to form a closed-loop control system, which realizes real-time monitoring and dynamic adjustment, so that the composite spectrum output by the light source module matches the target spectral data.

[0024] This embodiment solves the technical problem that traditional fixed spectra cannot adapt to the physiological characteristics of different leafy vegetable varieties by setting up a closed-loop control system consisting of a spectral sensing module, a spectral reconstruction unit, a spectral matching unit, and a spectral optimization unit. It achieves dynamic and precise spectral matching for different leafy vegetable varieties. The system can automatically adjust the luminous intensity of each LED unit according to the real-time collected spectral data, so that the output spectrum accurately matches the target preservation spectrum, thereby significantly improving the chlorophyll retention rate, effectively reducing weight loss, and extending shelf life.

[0025] In this preferred embodiment, the spectral sensing module is specifically a spectral sensor with eight visible light channels, and the multi-channel spectral data is the light intensity values ​​of the eight channels output by the spectral sensor. In this embodiment, the AS7341 spectral sensor is preferably used. This sensor has eight independent visible light channels, each covering a different spectral range. The F1 channel has a center wavelength of 415±5 nm and a bandwidth of approximately 40 nm, primarily detecting ultraviolet to blue light. The F2 channel has a center wavelength of 445±5 nm and a bandwidth of approximately 40 nm, responsible for blue light detection. The F3 channel has a center wavelength of 480±5 nm and a bandwidth of approximately 40 nm, covering the blue-green light band. The F4 and F5 channels have center wavelengths of 515±5 nm and 555±5 nm respectively, both with a bandwidth of 40 nm, jointly responsible for green light detection. The F6 channel has a center wavelength of 590±5 nm and a bandwidth of approximately 40 nm, corresponding to the yellow light region. The F7 channel has a center wavelength of 630±5 nm and a bandwidth of approximately 40 nm, used for orange light detection. The F8 channel has a center wavelength of 680±5 nm and a bandwidth of approximately 40 nm, specifically responsible for red light detection.

[0026] The sensor communicates with the main controller via an I²C interface, with a communication rate set to 100 kHz and an integration time configured to 50 milliseconds. Each channel can output a light intensity value with 12-bit precision.

[0027] The hardware implementation of this system is based on Figure 2 The connection diagram shown is the basis; In the diagram, 12VDC represents a 12-volt DC input power supply, providing energy to the entire system; VCC represents the positive terminal of the power supply, providing operating voltage to each chip and module; GND is the common ground terminal. The core of the control module uses STMicroelectronics' STM32F103RET6 as the main control chip. This chip is based on the ARM Cortex-M3 core, has a main frequency of 72MHz, and has 512KB Flash memory and 64KB SRAM, which is sufficient to meet the storage and operation requirements of neural network models and spectral databases.

[0028] Regarding signal connections, the chip pin labels, such as PA8 and PA9, are the general-purpose input / output pin numbers of this microcontroller. The system communicates with the spectral sensing module (using the AS7341 multi-channel spectral sensor chip) through its I²C1 interface (pins PC0-SCL as the serial clock line and PC1-SDA as the serial data line). Simultaneously, the main control chip generates five independent PWM signals (i.e., pulse width modulation signals, output from pins PA8-PA12, corresponding to PWM1 to PWM5 respectively) via timer TIM1 to precisely control five constant current drive circuits, thereby driving four monochromatic LEDs with center wavelengths of 400nm, 430nm, 451nm, and 645nm, as well as a white LED in the light source module.

[0029] The user interface module interacts with the main control chip via GPIO. Specifically, it connects to user buttons via GPIO pins (PB0-PB2) for selecting vegetable types, and drives a digital tube display to show the current status via specific pins.

[0030] Through the optimized design of the hardware architecture described above, the system achieves a balance between high performance and low cost. The ample storage space and computing power provided by the STM32F103RET6 main control chip ensure the stable operation of the neural network model and spectral database. Precise PWM output control, combined with the high-precision detection of the AS7341 spectral sensor, establishes a reliable closed-loop control foundation. The standardized communication interface design ensures efficient data interaction between modules, providing hardware support for the system's real-time performance and stability. This hardware configuration effectively controls manufacturing costs while ensuring the core performance of the system.

[0031] In this preferred embodiment, the spectral sensing module is specifically a spectral sensor with eight visible light channels, and the multi-channel spectral data is the light intensity value of the eight channels output by the spectral sensor; the multiple independently controllable LED units of different wavelengths in the light source module specifically include four monochromatic LEDs with center wavelengths of 400nm±5nm, 430nm±5nm, 451nm±5nm, and 645nm±5nm, and at least one white LED for providing a continuous background spectrum, and each LED unit is connected to the control module through an independent constant current driving circuit; In practice, the LED array consists of multiple monochromatic LED chips of different wavelengths and one white LED. The wavelength combination is specially designed based on the absorption characteristics of chlorophyll a and b. 400nm±5nm blue LED: This wavelength is at the edge of chlorophyll absorption and plays an important role in exciting photomorphogenesis reactions; 430nm±5nm blue LED: This wavelength can be effectively absorbed by chlorophyll and is an important supplement to the blue light region; 451nm±5nm blue LED: This wavelength band closely matches the maximum absorption peak of chlorophyll b, which is crucial for improving light energy utilization. 645nm±5nm red LED: This wavelength band is highly matched with the maximum absorption peak of chlorophyll a in the red light region, and is the core wavelength band driving photosynthesis; White LEDs: Used to provide a continuous background spectrum, fill the spectral valleys between monochromatic LEDs, improve the color rendering of the light source, and keep the color temperature of the synthesized light source in the warm white light range, thus serving both preservation and lighting functions.

[0032] Please see Figure 4To achieve a uniform spectral distribution, the LEDs on the substrate of the light source module are arranged in a concentric circle staggered arrangement. Specifically, LEDs of different wavelengths are arranged on multiple concentric rings, and LEDs of different wavelengths are arranged alternately on any ring and between adjacent rings, with a center-to-center distance of 20mm to 30mm between each LED.

[0033] In this embodiment, the light source module is implemented as follows: To achieve uniform illumination distribution and avoid local spectral unevenness caused by the aggregation of LEDs of the same color, the LEDs are arranged in a concentric circle staggered arrangement on the substrate. LEDs of different wavelengths are arranged on multiple concentric rings, and LEDs of different wavelengths are arranged alternately on the same ring and between adjacent rings; for example, a 645nm red LED is surrounded by 400nm, 430nm, 451nm or white LEDs, instead of another 645nm LED. This arrangement effectively reduces the self-absorption and spatial interference of photons of the same wavelength, and promotes the full mixing of photons of different wavelengths in the irradiation space, thereby forming a mixed light with extremely high spectral consistency on the target irradiation surface. The center-to-center spacing between LEDs is preferably 20mm to 30mm. This spacing can ensure the uniformity of light intensity while avoiding heat accumulation and optical crosstalk.

[0034] Each wavelength channel of LEDs is independently connected to a constant current drive circuit, which receives a pulse width modulation (PWM) signal from the control module to precisely regulate the current flowing through the LED group, thereby steplessly adjusting its luminous intensity (i.e., weighting). The above arrangement method achieves a highly uniform spectral distribution on the target irradiation surface, and the spectral characteristics of each irradiation point remain consistent, effectively ensuring the consistency and stability of light during the post-harvest preservation of leafy vegetables.

[0035] In a further embodiment, the target spectral data in the optimal preservation spectral database is obtained through the following method: The optimal preservation spectrum for each leafy vegetable variety was determined based on light experiments and post-harvest quality analysis of different leafy vegetable samples. The light experiments included applying irradiation with different spectral energy distributions to sample groups and monitoring multiple post-harvest quality indicators at multiple time points. By conducting statistical analysis and multi-objective optimization on the multiple post-harvest quality indicators, the spectral energy distribution curve that achieves the best overall preservation effect is determined. The spectral energy distribution curve forms energy peaks in the blue and red light regions corresponding to the chlorophyll absorption peaks, and suppresses the energy ratio of the far-red light band. The spectral energy distribution curve is discretized into a predetermined number of data points in the wavelength range of 380-780nm to form target spectral data for each leafy vegetable species; the multiple post-harvest quality indicators include chlorophyll content, weight loss rate and sensory quality indicators.

[0036] In this embodiment, the establishment of the optimal preservation spectrum database was accomplished through systematic experimental design and data analysis: First, fresh, disease-free samples of 10 common leafy green vegetables, such as spinach and lettuce, with uniform maturity were selected as experimental materials. The samples were pre-cooled immediately after harvesting and the experiment was carried out under uniform temperature and humidity conditions to eliminate the interference of environmental variables. The experimental design divided each vegetable sample into multiple experimental groups and a dark control group. A high-precision programmable LED light source system was used to apply spectral energy distributions consisting of monochromatic light at 400nm, 430nm, 451nm, and 645nm and different combinations of light intensities to each experimental group. An intermittent irradiation mode of 12 hours of light / 12 hours of darkness was adopted. At the set time points of days 1, 3, 5, and 7, the system measured and recorded the quality indicators of each experimental group, including chlorophyll content measured using a chlorophyll meter, weight loss rate calculated by weighing, and sensory evaluation based on color, wilting degree, and area of ​​decay.

[0037] Through statistical analysis of experimental data, an analysis of variance method was used to establish a correlation model between different spectral parameters and the freshness rate of various quality indicators. A multi-objective optimization algorithm was then used to solve for the spectral energy distribution curve that can best maintain chlorophyll content and overall quality.

[0038] The obtained optimal preservation spectrum exhibits significant energy peaks in the blue light region (430-451 nm) and the red light region (645-660 nm). These bands highly coincide with the maximum absorption peaks of chlorophyll a and b, which helps to maximize light energy absorption efficiency and maintain chloroplast structural stability. In addition to the main absorption peaks, the spectrum contains continuous and balanced broadband background light to fill the spectral valleys of single wavelengths, while providing necessary triggering signals for photoreceptors related to preservation, such as phytochromes and cryptochromes. In particular, a low energy ratio is maintained in the far-red light band (700-750 nm) to avoid activating physiological signals that accelerate aging. By minimizing the proportion of far-red light, the system can help vegetables maintain a youthful state with slow physiological activity. Finally, this continuous spectral curve is discretized and sampled at a resolution of 2.5 nm in the wavelength range of 380-780 nm, and the resulting spectral data sequence of 160 data points is stored in the database as the target spectrum.

[0039] The optimal preservation spectral database established through the aforementioned systematic experimental design and data analysis methods, based on in-depth research into the physiological characteristics of different leafy vegetables, shows that the energy peaks in the blue and red light regions are highly matched with chlorophyll absorption characteristics, effectively promoting efficient light energy utilization. Simultaneously, by suppressing the energy ratio in the far-red light band, the physiological activity of vegetables is cleverly regulated, delaying the aging process. This spectral energy distribution, obtained based on a multi-objective optimization algorithm, comprehensively considers multiple preservation indicators such as chlorophyll retention, moisture retention, and sensory quality, ensuring the best overall preservation effect. The discretized spectral data sequence provides a reliable data foundation for the precise control of the system, enabling different types of leafy vegetables to obtain light conditions most suitable for their physiological characteristics, thereby significantly improving preservation effects and extending shelf life.

[0040] In a further embodiment, the spectral reconstruction model built into the spectral reconstruction unit is specifically a neural network model. This model takes the spectral data of the 8 channels output by the spectral sensing module as input and maps it to generate high-resolution real-time spectral data with 160 data points. The 160 data points are arranged in wavelength order, uniformly covering the visible light band from 380nm to 780nm, and each data point represents its normalized light intensity value at the corresponding wavelength.

[0041] The neural network model specifically adopts a multilayer perceptron structure, with an input layer of 8 nodes, each corresponding to one of the 8 channels of the spectral sensing module; and an output layer of 160 nodes. The output values ​​of the 160 nodes sequentially constitute a high-resolution real-time spectral sequence covering the 380nm to 780nm band and having a spectral resolution of 2.5nm.

[0042] The neural network model is constructed through the following process:

[0043] Training data is collected based on multiple mixed spectral samples generated by the light source module. Different spectra are generated by adjusting the PWM weight combination of each LED unit, and the spectral data of the 8 channels of the spectral sensing module and the high-resolution true spectral data as the training target are collected simultaneously.

[0044] The multilayer perceptron structure is used for model training, with the collected spectral data from 8 channels as input and the corresponding high-resolution spectral ground truth data as the training target. The trained neural network model is output. After format conversion and quantization, the neural network model is integrated into the embedded system of the control module and run to reconstruct the spectral data of the 8 channels of the spectral sensing module into a 160-point high-resolution spectrum in real time.

[0045] In this embodiment, the specific construction process of the neural network model is as follows: Training data acquisition: Using the light source module of this invention, which contains five LED chips with different center wavelengths (400nm, 430nm, 451nm, 645nm, white light), 10,000 different PWM weight combinations [W1, W2, W3, W4, W5] (each weight uniformly distributed in the range of 0-1) are randomly generated to drive the light source to produce different mixed spectra. Under each weight combination, the readings of eight channels of the AS7341 sensor and 160-point spectral data (380-780nm, resolution 2.5nm) acquired by the high-precision spectrometer Ocean Optics USB4000 are recorded simultaneously. Data preprocessing: Divide the 8-channel readings of AS7341 by 4095 (maximum value of 12-bit ADC), divide the 160-point data of the spectrometer by the maximum value of each channel, and normalize all data to the [0,1] interval; divide the 10,000 sets of data into training set, validation set and test set in a ratio of 7000:1500:1500. Model Training: The MLP model is trained on a PC using the PyTorch framework, with the following parameters: Input layer: 8 nodes; Hidden layer 1: 128 neurons, ReLU activation function. This layer is used to learn complex nonlinear features from an 8-dimensional input. Hidden layer 2: 64 neurons, ReLU activation function. This layer is used to further combine and abstract features in preparation for outputting high-resolution spectra. Output layer: 160 neurons, sigmoid activation function, each neuron outputs a value between 0 and 1, corresponding to the normalized intensity of the reconstructed spectrum at 160 wavelength points; Loss function: Mean squared error (MSE), used to effectively penalize the difference between the predicted spectrum and the true spectrum; Optimizer: Adam, learning rate 0.001; Batch size: 32, Number of training rounds: 100; Early stopping technique was used during training. Training was terminated when the loss on the validation set no longer decreased for 10 consecutive rounds. The final model performance on the test set achieved a root mean square error (RMSE) of <0.015 and a coefficient of determination (R²) of >0.99. Model deployment: The trained model is quantized into INT8 format using TensorFlow Lite for Microcontrollers, converted into a C language array, and integrated into the STM32 project; in the embedded system, the inference time for spectral reconstruction is less than 50ms.

[0046] In this preferred embodiment, the spectral matching unit is configured to use a coefficient of determination R² as a quantitative index of the matching degree, and R² is calculated using the following formula: ; Where, represents the spectral intensity value of the target spectrum at the i-th wavelength point, represents the spectral intensity value of the reconstructed real-time spectrum at the i-th wavelength point, represents the average intensity of the target spectrum at all wavelength points, and i is the wavelength point index, with a value range of 1 to 160; The preset threshold is set to 98%. When R² ≥ 98%, the spectrum is considered to be well matched. When R² < 98%, the spectrum optimization unit is triggered. The spectral optimization unit is configured to use the least squares method as the optimization algorithm, and its objective function is: minΣ[(λ) - (λ)]²; Where λ is the wavelength variable, (λ) is the target spectral intensity at wavelength λ, and (λ) is the real-time spectral intensity at wavelength λ. The optimization variables are the weight coefficients W1, W2, W3, W4, and W5 of the 5 LED units, where each weight coefficient Wi corresponds to the luminous intensity weight of one LED. The constraint condition is that for all i, 0 ≤ Wi ≤ 1, and the weighting coefficient Wi is proportional to the PWM duty cycle driving the LED. The spectral optimization unit is further configured to convert the optimal weight coefficient set {W1, W2, W3, W4, W5} obtained by optimization calculation into corresponding PWM duty cycle signals and output them to each LED driving circuit of the light source module.

[0047] The specific workflow of the closed-loop control system is as follows: after the spectral optimization unit outputs new driving parameters to the light source module, the control module starts a 100-500ms delay program to wait for the output spectrum to reach a stable state. After the delay ends, the control module controls the spectral sensing module to perform a new round of spectral data acquisition; The collected spectral data are sequentially transmitted to the spectral reconstruction unit for high-resolution spectral reconstruction, and then transmitted to the spectral matching unit for matching degree verification with the target spectrum; When the verification result shows that the matching degree has not reached the preset threshold, the control module calls the spectral optimization unit again to perform iterative calculation and generate updated driving parameters; This process is repeated until the matching degree output by the spectral matching unit reaches or exceeds a preset threshold, forming a complete closed-loop feedback optimization mechanism.

[0048] like Figure 3As shown, the complete control logic from spectral acquisition, reconstruction, matching to optimization is illustrated, specifically demonstrating the timing relationships and data processing flow of each functional unit: Initialization Phase: After the system is powered on, initialization operations are performed, including peripheral initialization and parameter settings. First, the I²C bus, PWM output, GPIO interface and digital tube display are initialized and configured. Then, a database consisting of 160 spectral data points of 10 kinds of vegetables pre-stored in the Flash memory is loaded, and a pre-trained neural network model is loaded. The key operating parameters are set as follows: spectral matching degree threshold is 0.98, spectral sampling period is 3600 seconds (1 hour), and the stable delay after spectral optimization is 300 milliseconds. Spectral matching analysis implementation: The coefficient of determination R² is used as the quantitative index of spectral matching degree. The matching degree calculation process is as follows: First, the mean intensity of the target spectrum at all wavelengths is calculated. Then, the sum of squares of the residuals between the reconstructed real-time spectrum and the target spectrum, as well as the total sum of squares of the target spectrum, are calculated. Finally, the results are obtained by... The formula yields the matching degree result, and the calculation covers all 160 wavelength points to ensure the comprehensiveness and accuracy of the assessment.

[0049] Spectral optimization implementation: When the spectral matching degree does not reach the preset threshold, the system starts the least squares method to optimize the spectrum. The optimization objective is to minimize the sum of squares of the differences between the reconstructed real-time spectrum and the target spectrum at all wavelengths. The optimization variables are the weight coefficients W1 to W5 of the 5 LED units. Each weight coefficient must meet the constraint condition of 0 to 1.

[0050] The system uses the Levenberg-Marquardt algorithm for iterative solution. The optimization process first initializes the weight coefficients, then calculates the error gradient under the current weights in each iteration, updates the weight coefficients according to the gradient direction and applies the constraints, and then calculates the new error value. This process is repeated until the error is less than the set tolerance or the maximum number of iterations is reached.

[0051] After optimization, the system converts the obtained weight coefficients into PWM control signals. Each weight coefficient is linearly mapped to a PWM duty cycle value in the range of 0-1000, corresponding to a brightness adjustment range of 0%-100%. Finally, the system achieves precise control of the luminous intensity through the independent constant current drive circuit of each LED unit.

[0052] The system forms a complete closed-loop control: after the spectral optimization unit outputs new driving parameters, the control module starts a delay program to wait for the output spectrum to stabilize, and then re-triggers spectral acquisition and reconstruction, and performs matching degree verification. If the matching degree still does not meet the requirements, the optimization calculation is restarted. This process is repeated until the preset spectral matching standard is reached.

[0053] In this preferred embodiment, the user interface module includes: The selection unit includes a selection button for cyclically selecting vegetable types and a confirmation button for confirming the selection. The display unit includes a digital tube display for displaying the number of the currently selected vegetable type; The status indicator unit includes a dual-color LED indicator for indicating the system's operating status; The control module responds to the instruction input by the user through the selection unit by retrieving the target spectral data corresponding to the selected vegetable type number from the optimal preservation spectral database. The system firmware pre-stores optimal preservation spectral data for at least 10 kinds of leafy green vegetables, with each vegetable corresponding to a unique number. The control module calls up the corresponding target spectral data according to the number selected by the user.

[0054] In this embodiment, the user interface is implemented as follows: Selection Unit: "Up / Down" Button: Implemented via GPIO interrupt, supports long press for quick switching; "Confirm" button: Triggers target spectrum retrieval; Display unit: Two-digit LED display: Displays the currently selected vegetable number (01-10); Display encoding: BCD encoding is used, driven by 74HC595; Status indication unit: Green LED always on: good spectral matching (R² ≥ 0.98); Red LED constantly on: Spectrum optimization in progress; Red LED flashing (1Hz): System malfunction.

[0055] To verify the actual effect of the present invention, a systematic preservation effect test was conducted under controlled environmental conditions (temperature 4°C, relative humidity 90%, photoperiod 12 hours of light / 12 hours of darkness).

[0056] like Figure 5 As shown, the differences between the system of this invention and the traditional fixed spectrum system in terms of post-harvest preservation performance of leafy vegetables are intuitively demonstrated through comparative data. The left figure shows the comparison curve of chlorophyll retention rate over time in the post-harvest preservation process of leafy vegetables by the spectral optimization system of this invention and the traditional fixed spectrum system; the right figure shows the comparison curve of weight loss rate over time under the same conditions for the two systems. The test used lettuce as a representative sample and conducted comparative experiments using the spectral optimization system of this invention and the traditional fixed red-blue spectrum LED preservation system (red-blue light ratio 4:1) respectively. The changes in chlorophyll retention rate and weight loss rate were monitored daily for 7 consecutive days.

[0057] Experimental results show that the system of the present invention has significant advantages in several aspects: In terms of preservation effect, the chlorophyll retention rate curve of the system of the present invention changes more gently, indicating that the chlorophyll degradation process is effectively inhibited, and vegetables can maintain a good bright green appearance throughout the entire shelf life; at the same time, the system significantly reduces the intensity of transpiration by optimizing the spectrum to regulate the physiological activities of vegetables, which significantly slows down the rate of water loss and allows vegetables to maintain a plump and upright appearance for a long time.

[0058] Regarding shelf life, according to industry standards, vegetables lose their commercial value when chlorophyll retention drops below 60% or weight loss exceeds 12%. Experimental data shows that traditional fixed-spectrum systems reach the end of their shelf life around day 5, while the system of this invention extends the shelf life to over day 9, significantly enhancing the commercial value of the product. This result fully demonstrates the superior performance and practical value of the system of this invention in post-harvest preservation of leafy vegetables.

[0059] In addition, the system of this invention also shows significant advantages in terms of system versatility, long-term operational stability and cost control, providing an efficient, reliable and economical solution for post-harvest preservation of leafy vegetables.

[0060] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. A spectral optimization system for postharvest preservation of leafy vegetables, characterized in that, include: The light source module consists of multiple independently controllable LED units of different wavelengths, used to generate a composite spectrum for irradiating leafy vegetables; The spectral sensing module is directly aligned with the light-emitting surface of the light source module, and is used to acquire the composite spectrum in real time and output multi-channel spectral data. The control module is connected to the spectral sensing module and the light source module; An optimal preservation spectral database is integrated into the memory of the control module, storing target spectral data corresponding to different leafy vegetable varieties; The control module includes interconnected functional units: The spectral reconstruction unit is used to receive multi-channel spectral data output by the spectral sensing module and generate high-resolution real-time spectra through the built-in spectral reconstruction model. A spectral matching unit is used to receive the high-resolution real-time spectrum output by the spectral reconstruction unit and perform a matching degree analysis with the target spectral data retrieved from the optimal preservation spectral database. The spectral optimization unit is used to calculate the optimal driving parameters of each LED unit through an optimization algorithm when the matching degree output by the spectral matching unit is lower than a preset threshold, and output the driving parameters to the light source module. The spectral sensing module, the spectral reconstruction unit, the spectral matching unit, the spectral optimization unit in the control module, and the light source module are connected in sequence to form a closed-loop control system, which enables the composite spectrum output by the light source module to match the target spectral data.

2. The spectral optimization system for post-harvest preservation of leafy vegetables according to claim 1, characterized in that, The spectral sensing module is specifically a spectral sensor with 8 visible light channels, and the multi-channel spectral data is the light intensity value of the 8 channels output by the spectral sensor. The light source module contains multiple independently controllable LED units with different wavelengths, including four monochromatic LEDs with center wavelengths of 400nm±5nm, 430nm±5nm, 451nm±5nm, and 645nm±5nm, and at least one white LED for providing a continuous background spectrum. Each LED unit is connected to the control module through an independent constant current drive circuit.

3. The spectral optimization system for post-harvest preservation of leafy vegetables according to claim 2, characterized in that, To achieve a uniform spectral distribution, the LEDs on the substrate of the light source module are arranged in a concentric circle staggered arrangement. Specifically, LEDs of different wavelengths are arranged on multiple concentric rings, and LEDs of different wavelengths are arranged alternately on any ring and between adjacent rings, with a center-to-center distance of 20mm to 30mm between each LED.

4. The spectral optimization system for post-harvest preservation of leafy vegetables according to claim 1, characterized in that, The spectral reconstruction model built into the spectral reconstruction unit is a neural network model. This model takes the spectral data of the 8 channels output by the spectral sensing module as input and maps it to generate high-resolution real-time spectral data with 160 data points. The 160 data points are arranged in wavelength order, uniformly covering the visible light band from 380nm to 780nm, and each data point represents its normalized light intensity value at the corresponding wavelength.

5. A spectral optimization system for post-harvest preservation of leafy vegetables according to claim 4, characterized in that, The neural network model specifically adopts a multilayer perceptron structure, with an input layer of 8 nodes, each corresponding to one of the 8 channels of the spectral sensing module; and an output layer of 160 nodes. The output values ​​of the 160 nodes sequentially constitute a high-resolution real-time spectral sequence covering the 380nm to 780nm band and having a spectral resolution of 2.5nm.

6. A spectral optimization system for post-harvest preservation of leafy vegetables according to claim 1, characterized in that, The target spectral data in the optimal preservation spectral database are obtained through the following methods: The optimal preservation spectrum for each leafy vegetable variety was determined based on light experiments and post-harvest quality analysis of different leafy vegetable samples. The illumination experiment includes applying different spectral energy distributions to sample groups and monitoring multiple post-harvest quality indicators at multiple time points. By conducting statistical analysis and multi-objective optimization on the multiple post-harvest quality indicators, the spectral energy distribution curve that achieves the best overall preservation effect is determined. The spectral energy distribution curve forms energy peaks in the blue and red light regions corresponding to the chlorophyll absorption peaks, and suppresses the energy ratio of the far-red light band. The spectral energy distribution curve is discretized into a predetermined number of data points in the wavelength range of 380-780nm to form target spectral data for each leafy vegetable species. The postharvest quality indicators include chlorophyll content, weight loss rate, and sensory quality indicators.

7. A spectral optimization system for post-harvest preservation of leafy vegetables according to claim 5, characterized in that, The neural network model is constructed through the following process: Training data is collected based on multiple mixed spectral samples generated by the light source module. Different spectra are generated by adjusting the PWM weight combination of each LED unit, and the spectral data of the 8 channels of the spectral sensing module and the high-resolution true spectral data as the training target are collected simultaneously. The multilayer perceptron structure is used for model training, with the collected spectral data from 8 channels as input and the corresponding high-resolution spectral ground truth data as the training target. The trained neural network model is output. After format conversion and quantization, the neural network model is integrated into the embedded system of the control module and run to reconstruct the spectral data of the 8 channels of the spectral sensing module into a 160-point high-resolution spectrum in real time.

8. A spectral optimization system for post-harvest preservation of leafy vegetables according to claim 1, characterized in that, The spectral matching unit is configured to use a coefficient of determination R² as a quantitative index of the matching degree, which is calculated using the following formula: ； in, This represents the spectral intensity value of the target spectrum at the i-th wavelength. This represents the spectral intensity value at the i-th wavelength point in the reconstructed real-time spectrum. This represents the average intensity of the target spectrum across all wavelengths, where i is the wavelength index, ranging from 1 to 160. The preset threshold is set to 98%. When R² ≥ 98%, the spectrum is considered to be well matched. When R² < 98%, the spectrum optimization unit is triggered. The spectral optimization unit is configured to use the least squares method as the optimization algorithm, and its objective function is: minΣ[ (l) - (l)]² Where λ is the wavelength variable. (λ) represents the target spectral intensity at wavelength λ. (λ) represents the real-time spectral intensity at wavelength λ; The optimization variables are the weight coefficients W1, W2, W3, W4, and W5 of the 5 LED units, where each weight coefficient Wi corresponds to the luminous intensity weight of one LED. The constraint condition is that for all i, 0 ≤ Wi ≤ 1, and the weighting coefficient Wi is proportional to the PWM duty cycle driving the LED. The spectral optimization unit is further configured to convert the optimal weight coefficient set {W1, W2, W3, W4, W5} obtained by optimization calculation into corresponding PWM duty cycle signals and output them to each LED driving circuit of the light source module.

9. A spectral optimization system for post-harvest preservation of leafy vegetables according to claim 1, characterized in that, The specific workflow of the closed-loop control system is as follows: after the spectral optimization unit outputs new driving parameters to the light source module, the control module starts a 100-500ms delay program to wait for the output spectrum to reach a stable state. After the delay ends, the control module controls the spectral sensing module to perform a new round of spectral data acquisition; The collected spectral data are sequentially transmitted to the spectral reconstruction unit for high-resolution spectral reconstruction, and then transmitted to the spectral matching unit for matching degree verification with the target spectrum; When the verification result shows that the matching degree has not reached the preset threshold, the control module calls the spectral optimization unit again to perform iterative calculation and generate updated driving parameters; This process is repeated until the matching degree output by the spectral matching unit reaches or exceeds a preset threshold, forming a complete closed-loop feedback optimization mechanism.

10. A spectral optimization system for post-harvest preservation of leafy vegetables according to claim 1, characterized in that, It also includes a user interface module, which includes: The selection unit includes a selection button for cyclically selecting vegetable types and a confirmation button for confirming the selection. The display unit includes a digital tube display for displaying the number of the currently selected vegetable type; The status indicator unit includes a dual-color LED indicator for indicating the system's operating status; The control module responds to the instruction input by the user through the selection unit by retrieving the target spectral data corresponding to the selected vegetable type number from the optimal preservation spectral database. The system firmware pre-stores optimal preservation spectral data for at least 10 kinds of leafy green vegetables, with each vegetable corresponding to a unique number. The control module calls up the corresponding target spectral data according to the number selected by the user.

Images