Communication method for fresh agricultural product detection data

Abstract

The invention relates to the technical field of agricultural Internet of Things and intelligent sensing, in particular to a communication method for fresh agricultural product detection data, comprising: acquiring first detection data of a fresh agricultural product, the first detection data comprising a spectral signal; preprocessing the first detection data to obtain second detection data; performing credibility screening processing on the second detection data to obtain third detection data and the credibility of the third detection data; and selecting a transmission strategy based on the third detection data and the credibility of the third detection data, and transmitting the third detection data based on the transmission strategy. Based on the technical scheme of the invention, the problem of contradiction between detection data redundancy and transmission efficiency is effectively solved.

Description

Technical Field

[0001] The present invention relates to the field of agricultural Internet of Things and intelligent sensing technologies, and particularly to a communication method for fresh agricultural product detection data. Background Art

[0002] Since the development of near-infrared spectroscopy technology in the 1950s, with its advantages of high speed, non-destructive, and multi-index synchronous detection, it has been widely used in fields such as material analysis, environmental monitoring, and biomedicine. With the progress of computer technology, the hardware performance and signal processing capabilities of spectroscopic instruments have been continuously improved, enabling the gradual deepening of the application of near-infrared technology in scenarios such as food quality control and agricultural detection.

[0003] Current quality detection equipment for fresh agricultural products such as fruits and vegetables mainly relies on near-infrared spectroscopy sensors with a fixed sampling rate for data acquisition. However, in a conventional detection environment, the fixed sampling method results in more than 30% redundant data, which not only increases the burden of data storage and transmission but also exacerbates the energy consumption of the equipment. In addition, existing equipment usually operates in a continuous high-power mode, with a battery life of less than 72 hours, making it difficult to meet the requirements of long-term online monitoring. After the signal is transmitted to the equipment, the existing data acquisition terminal lacks an effective end-cloud collaborative data recovery mechanism, resulting in a data loss rate of up to 12% - 18% and a data reconstruction error (NRMSE) of more than 10% when the communication is interrupted, affecting the detection accuracy. Summary of the Invention

[0004] In view of the above problems, the present application provides a communication method for fresh agricultural product detection data, aiming to solve the contradiction between redundant detection data and transmission efficiency.

[0005] An embodiment of the present application provides a communication method for fresh agricultural product detection data, including:

[0006] Obtaining first detection data of the fresh agricultural product, where the first detection data includes spectral signals;

[0007] Preprocessing the first detection data to obtain second detection data;

[0008] Performing a credibility screening process on the second detection data to obtain third detection data and the credibility of the third detection data;

[0009] Selecting a transmission strategy based on the third detection data and the credibility of the third detection data, and transmitting the third detection data based on the transmission strategy.

[0010] In the above embodiments, through spectral signal acquisition, data preprocessing, credibility screening, and dynamic transmission strategy selection, the efficient and reliable transmission of fresh agricultural product detection data is achieved. Preprocessing and credibility screening improve data quality, and the transmission strategy based on credibility optimizes network resource utilization, which is particularly suitable for the quality monitoring scenario of fresh agricultural products in low-bandwidth or high-noise environments.

[0011] In some possible implementation manners, the obtaining of the first detection data of the fresh agricultural product includes:

[0012] Collecting the first detection data of the fresh agricultural product;

[0013] Analyzing the time-domain change characteristics of the first detection data, and adjusting the sampling rate of the first detection data based on a dynamic adaptive downsampling algorithm for the time-domain change characteristics.

[0014] In the above embodiments, by analyzing the time-domain change characteristics through a dynamic adaptive downsampling algorithm, the data volume is significantly reduced on the premise of ensuring that key information is not lost, solving the problem of redundant transmission caused by the traditional fixed sampling rate, and is particularly suitable for data acquisition scenarios with significant time-varying characteristics such as the spectral signals of fresh agricultural products.

[0015] In some possible implementation manners, the obtaining of the first detection data of the fresh agricultural product includes:

[0016] Collecting the first detection data of the fresh agricultural product;

[0017] Analyzing the historical data time-series correlation of the first detection data based on a prediction model of a long short-term memory network to generate a dynamic sampling rate, and updating the sampling rate of the first detection data based on the dynamic sampling rate.

[0018] In the above embodiments, using the LSTM network prediction model to achieve dynamic sampling rate adjustment can capture the time-series correlation of historical data, effectively balancing data integrity and transmission efficiency. This technology is particularly suitable for agricultural product detection data with periodic change rules, significantly reducing the generation of invalid data compared with static sampling methods.

[0019] In some possible implementation manners, the preprocessing of the first detection data to obtain the second detection data includes:

[0020] Performing noise reduction processing on the first detection data based on a wavelet denoising algorithm to obtain a first denoised detection data, where the denoising parameters of the wavelet denoising algorithm are dynamically adjusted;

[0021] Performing filtering processing on the first denoised detection data based on a low-pass filter to obtain the second detection data.

[0022] In the above - mentioned embodiment, a two - stage noise suppression mechanism is formed by cascading wavelet denoising with dynamic parameters and low - pass filtering: wavelet denoising is aimed at high - frequency random noise, and low - pass filtering eliminates low - frequency baseline drift. This combination significantly improves the signal - to - noise ratio of the spectral signal, providing a high - quality data basis for subsequent credibility calculation.

[0023] In some possible implementation manners, the pre - processing of the first detection data to obtain the second detection data includes:

[0024] Performing parallel processing on the first detection data based on a wavelet denoising algorithm and a low - pass filter, and obtaining the second detection data through weighted fusion of the results of the parallel processing;

[0025] Among them, the wavelet denoising algorithm is used to suppress high - frequency noise in the first detection data, and the low - pass filter is used to correct low - frequency baseline drift in the first detection data.

[0026] In the above - mentioned embodiment, the collaborative optimization of high - frequency noise suppression and low - frequency drift correction is realized through a parallel processing architecture, and the weighted fusion strategy retains the advantageous features of the two algorithms. Compared with serial processing, this method significantly improves the processing speed, and the spectral characteristics of the fused data are closer to the true values.

[0027] In some possible implementation manners, the credibility screening process of the second detection data to obtain the third detection data and the credibility of the third detection data includes:

[0028] Performing primary screening on the second detection data to obtain primary data; the primary screening process includes eliminating obviously invalid data;

[0029] Performing advanced screening on the primary data to obtain the third detection data; the advanced screening process includes constructing a normal distribution model based on cloud historical data, eliminating probability outliers, and dynamically updating the mean and standard deviation of the third detection data;

[0030] Performing dynamic weight calculation on the third detection data to obtain the credibility of the third detection data; the calculation of the credibility is based on a weighted combination of the signal - to - noise ratio and the spectral stability of the third detection data.

[0031] In the above - mentioned embodiment, a three - level credibility evaluation system of "primary screening - advanced screening - dynamic weight calculation" is constructed: primary screening quickly eliminates invalid data, intelligent outlier detection is realized based on the normal distribution model of cloud historical data, and the evaluation of the weighted combination of signal - to - noise ratio and spectral stability makes the credibility index more scientific.

[0032] In some possible implementation manners, the method of selecting the transmission strategy of the third detection data based on the third detection data and the credibility of the third detection data includes:

[0033] In response to the third detection data and the credibility of the third detection data satisfying a first preset condition, setting the transmission strategy of the third detection data to a deep sleep mode; wherein the deep sleep mode includes setting the detection data acquisition device to a low-power and low-frequency transmission mode;

[0034] In response to the third detection data and the credibility of the third detection data satisfying a second preset condition, setting the transmission strategy of the third detection data to an intermittent transmission mode; wherein the intermittent transmission mode includes performing compression processing on the third detection data through a compression algorithm;

[0035] In response to the third detection data and the credibility of the third detection data satisfying a third preset condition, setting the transmission strategy of the third detection data to a real-time upload mode.

[0036] In the above implementation manner, the proposed three-level transmission strategy (deep sleep / intermittent transmission / real-time upload) achieves an optimal balance between energy consumption and data integrity. The deep sleep mode can significantly reduce the power consumption of the terminal. The intermittent transmission mode greatly reduces the bandwidth occupation through the compression algorithm. The real-time upload mode ensures zero-latency transmission of key data.

[0037] In some possible implementation manners, the first preset condition is that the credibility of the third detection data is less than a first threshold within a first preset time, and it is statistically analyzed based on the mean value of the third detection data that the third detection data belongs to the data in the marginal area.

[0038] In some possible implementation manners, the second preset condition is that the credibility of the third detection data is greater than or equal to the first threshold and less than a second threshold within a second preset time, and it is statistically analyzed based on the mean value of the third detection data that the third detection data belongs to the data in the sub-marginal area.

[0039] In some possible implementation manners, the third preset condition is that the credibility of the third detection data is greater than or equal to the second threshold within a third preset time, and it is statistically analyzed based on the mean value of the third detection data that the third detection data belongs to the data in the core distribution area.

[0040] In the above some implementation manners, through the multi-dimensional condition judgment of statistical analysis and credibility thresholds, the data transmission levels are accurately divided. The data is divided into three categories: marginal area / sub-marginal area / core distribution area, making the selection of the transmission strategy have a clear mathematical basis and avoiding the misjudgment problem of traditional single-threshold determination.

[0041] In some possible implementation manners, the deep learning reconstruction model deployed based on the cloud reconstructs the lost packet data that appears during the transmission of the third detection data.

[0042] In the above implementation manner, the deep learning reconstruction model in the cloud solves the latency problem caused by the traditional retransmission mechanism, and can still achieve a data reconstruction accuracy of more than 95% under the condition of a packet loss rate of 20%, significantly improving the robustness of the data acquisition terminal for agricultural product quality monitoring.

[0043] In some possible implementation manners, the reconstruction of the lost packet data that appears during the transmission by the deep learning reconstruction model deployed based on the cloud includes:

[0044] Based on the incomplete data uploaded by the data acquisition terminal as the input of the deep learning reconstruction model, capture the periodic fluctuations and transient mutations of the incomplete data through a bidirectional LSTM network, extract spectral features using a 1D convolutional neural network, construct a reconstruction interpolation model for the lost packet data, and realize the reconstruction of the lost packet data;

[0045] Among them, the incomplete data is the data that has not lost packets in the third detection data, and the deep learning reconstruction model is a CNN-LSTM hybrid model.

[0046] In the above implementation manner, the CNN-LSTM hybrid model captures temporal features through a bidirectional LSTM and extracts spectral features through 1DCNN, and the reconstruction error of the incomplete data is less than 3%. This technology is particularly suitable for data reconstruction of spectral signals with both periodic and transient characteristics, and the reconstruction speed is increased by 2 times compared with a single model.

[0047] The above technical features can be combined in various suitable ways or replaced by equivalent technical features as long as the purpose of the present invention can be achieved.

[0048] Compared with the prior art, the beneficial effects of the present application are at least as follows:

[0049] (1) Significantly reduce the amount of sampled redundant data and increase the proportion of effective data. By dynamically adjusting the sampling rate for sampling, the proportion of effective data of the data acquisition terminal in a conventional detection environment is significantly increased. At the same time, based on the time-domain feature-based adaptive downsampling algorithm, the amount of redundant data and the hardware resource occupancy rate are significantly reduced, and the data redundancy problem in the fixed sampling mode is solved.

[0050] (2) Reduce transmission energy consumption and achieve ultra-low power operation. Through a three-level transmission strategy, the average power consumption of the data acquisition terminal in a low activity period is significantly reduced, the transmission success rate is significantly increased, and the single transmission delay is significantly shortened, significantly alleviating the problem of low energy efficiency of traditional communication protocols.

[0051] (3) Edge-cloud collaborative reconstruction mechanism to effectively recover lost data. On the data acquisition terminal side, compression technology and local cache queue are adopted. A CNN-LSTM hybrid model is deployed on the cloud side, and high-precision signal reconstruction is achieved through joint extraction of spatio-temporal features. Under extreme conditions of packet loss, the reconstruction error is greatly reduced, solving the problems of data loss and insufficient reconstruction accuracy caused by data transmission interruption. Description of the Drawings

[0052] In the following, the present invention will be described in more detail based on embodiments and with reference to the drawings. Among them:

[0053] Figure 1 is a schematic diagram of a communication method for fresh agricultural product detection data provided by an embodiment of the present application.

[0054] Figure 2 is a schematic diagram of a transmission strategy of a communication method for fresh agricultural product detection data provided by an embodiment of the present application. Detailed Embodiments

[0055] To make the above objects, features, and advantages of the present application more obvious and understandable, the present invention will be further described in detail below with reference to the drawings and specific embodiments. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of them. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present application belong to the scope of protection of the present application.

[0056] The terms used in the embodiments of the present application are only used to explain the specific embodiments of the present application, rather than to limit the present application.

[0057] There are three core defects in the existing agricultural product (such as fruits and vegetables) detection technology: First, the fixed sampling mode leads to a significant amount of redundant data and a low proportion of effective data; Second, due to the static transmission strategy and continuous high-power consumption operation of the traditional communication protocol, the battery life of the device is insufficient and the channel utilization rate is less than 40%; Third, in the scenario of data transmission interruption or packet loss, the existing technology relies on a single compressive sensing algorithm and has a high reconstruction error.

[0058] In view of this, to solve the technical problems existing in the prior art, see Figure 1 As shown, an embodiment of the present application discloses a communication method for fresh agricultural product detection data, including:

[0059] Step S11, obtaining first detection data of the fresh agricultural product, where the first detection data includes spectral signals.

[0060] Fresh agricultural products can be fruits and vegetables, livestock, poultry, aquatic products, etc. For example, they can be perishable products (strawberries, blueberries, cherries, spinach), and perishable products need to be monitored in real time for high moisture content; they can also be fruits and vegetables with rapid post-harvest quality changes (bananas, avocados, tomatoes), and their maturity can be detected by spectroscopy; they can also be freshness-sensitive products (salmon, beef, oysters), and protein / fat spoilage can be detected by near-infrared spectroscopy. In the embodiments of the present application, bananas in fruits and vegetables are mainly used as the test object for elaboration, but the specific object is not limited thereto.

[0061] The first detection data of fresh agricultural products can be collected by a data acquisition terminal, and the data acquisition terminal integrates multiple modules such as a data acquisition module, an edge processing unit, a communication optimization module, and a cloud recovery module. Among them, the data acquisition module integrates a near-infrared spectroscopy sensor with a working wavelength of 400 - 1600 nm and a resolution of 10 nm, which is used to obtain the spectral signal of fresh agricultural products in real time. The data acquisition module can also integrate other sensors such as accelerometer data and ambient temperature sensor data.

[0062] In some embodiments, the obtaining of the first detection data of the fresh agricultural products in step S11 includes:

[0063] Collect the first detection data of the fresh agricultural products;

[0064] Analyze the time-domain variation characteristics of the first detection data, and adjust the sampling rate of the first detection data based on the dynamic adaptive downsampling algorithm of the time-domain variation characteristics.

[0065] To meet the data acquisition requirements under different environmental conditions, a dynamic sampling rate adjustment module is also designed for the communication method of fresh agricultural product detection data. The dynamic sampling rate adjustment module analyzes the time-domain variation characteristics of the first detection data, and automatically adjusts the sampling frequency based on the change situation of the spectral signal by using the dynamic adaptive downsampling algorithm. Especially when the signal changes violently, the data acquisition terminal will automatically increase the sampling rate to obtain more refined data; while when the signal is stable, the data acquisition terminal will reduce the sampling frequency, thereby reducing the generation of redundant data and improving the efficiency and accuracy of data acquisition. Through this adaptive sampling strategy, the proportion of effective data is effectively increased. In the test, the proportion of effective data is increased from 68% in the traditional scheme to 92%.

[0066] As an alternative implementation of the dynamic sampling rate adjustment mechanism, in some other embodiments, the obtaining of the first detection data of the fresh agricultural products in step S11 can also include:

[0067] Collect the first detection data of the fresh agricultural products;

[0068] The prediction model based on the long short-term memory network analyzes the historical data time series correlation of the first detection data to generate a dynamic sampling rate, and updates the sampling rate of the first detection data based on the dynamic sampling rate.

[0069] In the above implementation, by replacing the dynamic sampling rate adjustment mechanism based on time domain feature parameters with a prediction model based on the long short-term memory network, a dynamic sampling rate instruction can be generated by analyzing the time series correlation of historical spectral signals. The model can further fuse the accelerometer data and ambient temperature sensor data of the data acquisition terminal as auxiliary inputs to achieve more accurate sampling control in specific scenarios.

[0070] Step S12: Preprocess the first detection data to obtain second detection data.

[0071] Data preprocessing is to transform the first detection data into second detection data more suitable for analysis through technical means such as data cleaning, data analysis, machine learning, and data mining.

[0072] In some embodiments, the preprocessing of the first detection data in step S12 to obtain second detection data includes:

[0073] Perform noise reduction processing on the first detection data based on the wavelet denoising algorithm to obtain first denoised detection data, where the denoising parameters of the wavelet denoising algorithm are dynamically adjusted;

[0074] Perform filtering processing on the first denoised detection data based on a low-pass filter to obtain second detection data.

[0075] Furthermore, in some embodiments, the data acquisition terminal is also integrated with a hybrid filtering module for preprocessing the first detection data.

[0076] The hybrid filtering module performs noise reduction processing on the first detection data based on the wavelet denoising algorithm. Specifically, the hybrid filtering module can use the Daubechies-4 wavelet basis function for 5-layer signal decomposition, and dynamically adjust the denoising coefficient according to the time domain change of the signal to suppress high-frequency noise. In addition, the wavelet basis function can also be replaced with the Symlet series to improve the decomposition accuracy of signal mutations by increasing the number of vanishing moments. Subsequently, the hybrid filtering module performs filtering processing on the first denoised detection data based on a low-pass filter to obtain second detection data. Specifically, a programmable FIR low-pass filter with an order of 64 and a cut-off frequency of 20 Hz can be used to eliminate low-frequency interference. In experimental tests, after the above processing, the signal-to-noise ratio (SNR) is increased from <30 dB of the traditional scheme to above 45 dB, significantly improving the signal quality.

[0077] As an alternative implementation of the hybrid filtering process, in some other embodiments, the preprocessing of the first detection data in step S12 to obtain the second detection data includes:

[0078] Performing parallel processing on the first detection data based on a wavelet denoising algorithm and a low-pass filter, and obtaining the second detection data through weighted fusion of the results of the parallel processing;

[0079] Among them, the wavelet denoising algorithm is used to suppress high-frequency noise in the first detection data, and the low-pass filter is used to correct low-frequency baseline drift in the first detection data.

[0080] In the above implementation, compared with the cascaded architecture of wavelet denoising and adjustable FIR filtering, through the parallel processing and weighted fusion architecture of wavelet denoising and FIR low-pass filtering, the targeted noise processing ability is enhanced, and at the same time, the parallel processing avoids error accumulation, and the calculation efficiency is greatly optimized.

[0081] Step S13, performing credibility screening processing on the second detection data to obtain the third detection data and the credibility of the third detection data.

[0082] The embodiments of the present application optimize the reliability of data transmission through a credibility screening mechanism, and further implement a dynamic hierarchical transmission strategy in combination with probability statistics methods.

[0083] In some embodiments, the performing credibility screening processing on the second detection data to obtain the third detection data and the credibility of the third detection data includes:

[0084] Performing primary screening processing on the second detection data to obtain primary data; the primary screening processing includes eliminating obviously invalid data;

[0085] Performing advanced screening processing on the primary data to obtain the third detection data; the advanced screening processing includes constructing a normal distribution model based on cloud historical data, eliminating probability outliers, and dynamically updating the mean and standard deviation of the third detection data;

[0086] Performing dynamic weight calculation on the third detection data to obtain the credibility of the third detection data; the calculation of the credibility is based on a weighted combination of the signal-to-noise ratio of the third detection data and the spectral stability of the third detection data.

[0087] In the above implementation, during the process of screening and processing the credibility of the second detection data, obvious invalid data is first removed through primary screening, such as the case where the signal amplitude is lower than the preset threshold or the spectral energy distribution is abnormal, to ensure the quality of the basic data. However, the advanced screening process further constructs a normal model (N(μ,σ2)) using historical cloud data, removes outliers with a probability of only 0.27% through the 3σ criterion (∣xi - μ∣> 3σ), and dynamically updates the mean and standard deviation to adapt to the real-time data stream. Finally, based on the dynamic weight formula, the data credibility is quantified, and through the weighted calculation of the signal-to-noise ratio (SNR) and spectral stability (Fstability), the data is mapped to different confidence intervals. Among them, high-credibility data (mean ± 1σ, covering about 68% of the normal data) is transmitted in real time, medium-credibility data (mean ± 1σ to ± 2σ, covering about 27%) is transmitted after batch compression, and low-credibility data (mean ± 2σ to ± 3σ, covering about 4.7%) triggers resampling or discarding. By deeply combining confidence interval division and probability statistics, while ensuring the real-time response of key data, the transmission frequency and power consumption of medium- and low-credibility data are significantly reduced, and finally the optimization goals of a comprehensive mistransmission rate ≤ 3% and a 60% reduction in energy consumption are achieved, providing highly reliable and low-power communication support for banana quality detection in a dynamic environment.

[0088] Optionally, the weighted calculation formula for the credibility is:

[0089] W = 0.6SNR + 0.4Fst

[0090] Where W is the credibility, SNR is the signal-to-noise ratio of the third detection data, and Fst is the spectral stability of the third detection data.

[0091] Step S14, select a transmission strategy based on the third detection data and the credibility of the third detection data, and transmit the third detection data based on the transmission strategy.

[0092] In the foregoing steps, the third detection data after screening and processing is marked with credibility, and further a transmission strategy is selected based on the third detection data and its data credibility, and the third detection data is transmitted based on the transmission strategy. Through the above implementation, the transmission strategy can be flexibly switched according to the reliability and transmission requirements of the data.

[0093] In some embodiments, the selecting the transmission strategy of the third detection data based on the third detection data and the credibility of the third detection data includes:

[0094] In response to the third detection data and the credibility of the third detection data satisfying a first preset condition, set the transmission policy of the third detection data to the deep sleep mode; wherein the deep sleep mode includes setting the detection data acquisition device to the low-power low-frequency transmission mode;

[0095] In response to the third detection data and the credibility of the third detection data satisfying a second preset condition, set the transmission policy of the third detection data to the intermittent transmission mode; wherein the intermittent transmission mode includes performing compression processing on the third detection data through a compression algorithm;

[0096] In response to the third detection data and the credibility of the third detection data satisfying a third preset condition, set the transmission policy of the third detection data to the real-time upload mode.

[0097] In the above implementation, as Figure 2 shown, in the low-power low-frequency transmission mode, the data acquisition terminal shuts down the power supply of non-core modules, reduces the power consumption to the limit of 0.3 mA, and only maintains the basic connection by sending a heartbeat packet (including the device status code) of 1 byte per minute. The transmission delay of the heartbeat packet is strictly controlled within 100 ms to ensure that the device can be remotely awakened. In the intermittent transmission mode, the third detection data is packed and sent after accumulating to 10 pieces, and is transmitted once every 2 - 5 seconds. The volume of the data packet is ≤ 20 B, and the credibility score and timestamp are embedded in the header to support cloud time series alignment. The power consumption of transmission processed by the compression algorithm in the intermittent transmission mode is about 5 mA. In the real-time upload mode, an improved CSMA / CA protocol is used to dynamically adjust the backoff window (the backoff time is shortened by 50% when the channel is idle), and the third detection data is directly transmitted to ensure high fidelity. In the real-time upload mode, the power consumption of real-time transmission of the third detection data is about 15 mA. The intermittent transmission mode includes performing compression processing on the third detection data through a compression algorithm, specifically, by retaining the key coefficients of the first 20% of the signal energy through the discrete cosine transform (DCT) compression algorithm to achieve a compression ratio of ≥ 80%.

[0098] In some embodiments, the first preset condition is: the credibility of the third detection data is less than a first threshold within a first preset time, and it is statistically analyzed based on the mean value of the third detection data that the third detection data belongs to the marginal area data.

[0099] Optionally, the first preset time can be 3 minutes, and the first threshold can be 0.5. The deep sleep mode is activated when the data acquisition terminal does not detect high-credibility data (credibility score W < 0.5) for 3 consecutive minutes and the data is in the interval of mean ± 2σ to ± 3σ.

[0100] In some embodiments, the second preset condition is that: within a second preset time, the credibility of the third detection data is greater than or equal to a first threshold and less than a second threshold, and based on the mean value of the third detection data, statistical analysis shows that the third detection data belongs to sub-marginal region data.

[0101] Optionally, the second preset time can be 3 minutes, the first threshold can be 0.5, and the second threshold can be 0.8. When the data acquisition terminal continuously detects medium-credibility data (credibility score 0.5 ≤ W < 0.8) for 3 minutes and the data is within the interval of mean ± 1σ to ± 2σ, the intermittent transmission mode is activated.

[0102] In some embodiments, the third preset condition is that: within a third preset time, the credibility of the third detection data is greater than or equal to the second threshold, and based on the mean value of the third detection data, statistical analysis shows that the third detection data belongs to core distribution region data.

[0103] Optionally, the third preset time can be 3 minutes, and the second threshold can be 0.8. When the data acquisition terminal continuously detects high-credibility data (credibility score W ≥ 0.8) for 3 minutes and the data is within the interval of mean ± 1σ, the real-time upload mode is activated.

[0104] Further in some embodiments, the communication method for fresh agricultural product detection data in the embodiments of the present application may further include step S15 shown by the dotted line in the appendix Figure 1 to reconstruct the lost packet data that appears during the transmission of the third detection data based on the deep learning reconstruction model deployed on the cloud.

[0105] The deep learning reconstruction model is an algorithm framework that uses a deep neural network to recover complete and high-quality information from partial, damaged, or low-quality data. Lost packet data is a data packet that fails to reach the destination successfully during network transmission. Based on the deep learning reconstruction model deployed on the cloud, high-quality reconstruction of the lost packet data that appears during the transmission of the third detection data can be achieved.

[0106] In some embodiments, the reconstruction of the lost packet data that appears during the transmission based on the deep learning reconstruction model deployed on the cloud includes:

[0107] Using the incomplete data uploaded by the data acquisition terminal as the input of the deep learning reconstruction model, capturing the periodic fluctuations and transient mutations of the incomplete data through a bidirectional LSTM network, extracting spectral features using a 1D convolutional neural network, and constructing a reconstruction interpolation model for the lost packet data to achieve the reconstruction of the lost packet data;

[0108] wherein, the incomplete data is the data of the third detection data that has not been lost, and the deep learning reconstruction model is a CNN-LSTM hybrid model.

[0109] In the above implementation, in the process of reconstructing the lost packet data, a CNN-LSTM hybrid model is adopted to achieve high-precision signal reconstruction. This deep learning reconstruction model takes the incomplete data uploaded by the data acquisition terminal, such as current signals (including timestamps, current values, and lost packet marks), as input. It captures the periodic fluctuations and transient mutations of the current signal through a bidirectional LSTM network, extracts spectral features using a 1D convolutional neural network (CNN), and constructs an interpolation model for reconstructing lost packet data. For example, when the data from the 5th to 6th seconds is lost, the model accurately interpolates the current curve for the missing period based on the upward trend from the 3rd to 4th seconds and the decay pattern from the 7th to 8th seconds. Among them, the CNN-LSTM hybrid model is a deep learning architecture that combines the advantages of convolutional neural networks (CNN) and long short-term memory networks (LSTM), and is widely used in spatio-temporal sequence data processing tasks.

[0110] Generally speaking, through the above technical solutions, the communication method for fresh agricultural product detection data in this application achieves the goals of reducing power consumption by 65% and increasing the effective data transmission rate by 80% in the agricultural scenario of banana quality detection, providing a complete solution with high precision and low power consumption for quality monitoring in the banana supply chain. Specifically, it has the following technical effects:

[0111] (1) Significantly reduce the amount of sampled redundant data and increase the proportion of effective data

[0112] By integrating a near-infrared spectroscopy sensor and dynamically adjusting the sampling rate, the proportion of effective data of the data acquisition terminal in a conventional detection environment is increased from 68% in the traditional solution to 92%. At the same time, based on the time-domain feature-based adaptive downsampling algorithm, the amount of redundant data is reduced by 60%, and the hardware resource occupancy rate is reduced by 40%, solving the problem of data redundancy in the fixed sampling mode.

[0113] (2) Optimize transmission energy consumption and achieve ultra-low power operation

[0114] Design a three-level sleep architecture and a dynamic priority marking algorithm, combined with an improved CSMA / CA protocol, to reduce the average power consumption of the device during the low activity period to 0.8 mA, a 96% reduction compared to the existing technology. The transmission success rate is increased from 75% to 93%, and the single transmission delay is shortened to 28 ms, significantly alleviating the problem of low energy efficiency of traditional communication protocols.

[0115] (3) End-cloud collaborative reconstruction mechanism to effectively recover lost data

[0116] On the data acquisition terminal side, DCT compression technology and a local cache queue are adopted, and a CNN-LSTM hybrid model is deployed in the cloud. High-precision signal reconstruction is achieved through joint extraction of spatio-temporal features. Under extreme conditions of 50% packet loss, the reconstruction error is stable at 4.2% - 5.1%, which is 58% higher than the traditional scheme, solving the problems of data loss and insufficient reconstruction accuracy caused by data transmission interruption.

[0117] Although the present invention has been described herein with reference to specific embodiments, it should be understood that these embodiments are merely examples of the principles and applications of the present invention. Therefore, it should be understood that many modifications can be made to the exemplary embodiments, and other arrangements can be designed, as long as they do not depart from the spirit and scope of the present invention as defined by the appended claims. It should be understood that the different dependent claims and the features described herein can be combined in a manner different from that described in the original claims. It should also be understood that the features described in connection with a single embodiment can be used in other described embodiments.

Claims

1. A communication method for fresh agricultural product detection data, characterized in that, Including: Obtaining first detection data of the fresh agricultural product, where the first detection data includes spectral signals; Preprocessing the first detection data to obtain second detection data; Performing credibility screening processing on the second detection data to obtain third detection data and the credibility of the third detection data; Selecting a transmission strategy based on the third detection data and the credibility of the third detection data, and transmitting the third detection data based on the transmission strategy.

2. The communication method for fresh agricultural product detection data according to claim 1, wherein The obtaining of the first detection data of the fresh agricultural product includes: Collecting the first detection data of the fresh agricultural product; Analyzing the time-domain change characteristics of the first detection data, and adjusting the sampling rate of the first detection data based on a dynamic adaptive downsampling algorithm for the time-domain change characteristics.

3. The communication method for fresh agricultural product detection data according to claim 1, characterized in that The obtaining of the first detection data of the fresh agricultural product includes: Collecting the first detection data of the fresh agricultural product; Analyzing the historical data time-series correlation of the first detection data based on a prediction model of a long short-term memory network to generate a dynamic sampling rate, and updating the sampling rate of the first detection data based on the dynamic sampling rate.

4. The communication method for fresh agricultural product detection data according to claim 1, characterized in that, The preprocessing of the first detection data to obtain second detection data includes: Performing noise reduction processing on the first detection data based on a wavelet denoising algorithm to obtain first denoised detection data, where the denoising parameters of the wavelet denoising algorithm are dynamically adjusted; Performing filtering processing on the first denoised detection data based on a low-pass filter to obtain second detection data.

5. The communication method for fresh agricultural product detection data according to claim 1, characterized in that, The preprocessing of the first detection data to obtain second detection data includes: Performing parallel processing on the first detection data based on a wavelet denoising algorithm and a low-pass filter, and obtaining second detection data through weighted fusion of the results of the parallel processing; Among them, the wavelet denoising algorithm is used to suppress high-frequency noise of the first detection data, and the low-pass filter is used to correct low-frequency baseline drift of the first detection data.

6. The communication method for fresh agricultural product detection data according to claim 1, wherein, The performing of credibility screening processing on the second detection data to obtain third detection data and the credibility of the third detection data includes: Performing primary screening processing on the second detection data to obtain primary data; the primary screening processing includes removing obviously invalid data; Performing advanced screening processing on the primary data to obtain third detection data; the advanced screening processing includes constructing a normal distribution model based on cloud historical data, removing probability outliers, and dynamically updating the mean and standard deviation of the third detection data; Performing dynamic weight calculation on the third detection data to obtain the credibility of the third detection data; the calculation of the credibility is based on a weighted combination of the signal-to-noise ratio of the third detection data and the spectral stability of the third detection data.

7. The communication method for fresh agricultural product detection data according to claim 1, characterized in that The selecting of the transmission strategy for the third detection data based on the third detection data and the credibility of the third detection data includes: In response to the third detection data and the credibility of the third detection data satisfying a first preset condition, setting the transmission strategy of the third detection data to a deep sleep mode; where the deep sleep mode includes setting the detection data acquisition device to a low-power low-frequency transmission mode; In response to the third detection data and the credibility of the third detection data satisfying a second preset condition, set the transmission strategy of the third detection data to an intermittent transmission mode; wherein the intermittent transmission mode includes compressing the third detection data through a compression algorithm; In response to the third detection data and the credibility of the third detection data satisfying a third preset condition, set the transmission strategy of the third detection data to a real-time upload mode.

8. The communication method for fresh agricultural product detection data according to claim 7, wherein The first preset condition is that the credibility of the third detection data is less than a first threshold within a first preset time, and it is statistically analyzed based on the mean value of the third detection data that the third detection data belongs to marginal area data; The second preset condition is that the credibility of the third detection data is greater than or equal to the first threshold and less than a second threshold within a second preset time, and it is statistically analyzed based on the mean value of the third detection data that the third detection data belongs to sub-marginal area data; The third preset condition is that the credibility of the third detection data is greater than or equal to the second threshold within a third preset time, and it is statistically analyzed based on the mean value of the third detection data that the third detection data belongs to core distribution area data.

9. The communication method for fresh agricultural product detection data according to claim 1, characterized in that, It further includes: Reconstructing the lost packet data that appears during the transmission of the third detection data based on a deep learning reconstruction model deployed in the cloud.

10. The communication method for fresh agricultural product detection data according to claim 9, characterized in that, The reconstructing of the lost packet data that appears during the transmission by the deep learning reconstruction model deployed in the cloud includes: Using the incomplete data uploaded by the data acquisition terminal as the input of the deep learning reconstruction model, capturing the periodic fluctuations and transient mutations of the incomplete data through a bidirectional LSTM network, extracting spectral features using a 1D convolutional neural network, and constructing a reconstruction interpolation model for the lost packet data to achieve the reconstruction of the lost packet data; wherein, the incomplete data is the data in the third detection data that has not been lost, and the deep learning reconstruction model is a CNN-LSTM hybrid model.

Images