Intelligent method and system for tamarind pulp processing monitoring

Abstract

The invention relates to the technical field of artificial intelligence, discloses an intelligent method and system for tamarind pulp processing monitoring, and aims to solve the problems that pulp coking and water residue exceeding standards are caused by fixed parameters in an existing drying process, the subjectivity of manual detection is high, and a traditional visual model is difficult to meet the industrial real-time performance. According to the method, visible light and near-infrared images of tamarind pulp are synchronously obtained through dual-band spectral imaging, surface texture roughness and internal light transmittance indexes are extracted respectively and fused to generate a maturity index, and the drying temperature is dynamically regulated and controlled according to the maturity index. The system comprises a dual-band image acquisition module, an edge calculation processing module and a temperature control module, and supports real-time detection and closed-loop regulation and control at an industrial edge end. According to the method, high precision, self-adaption and intelligence of the tamarind pulp drying process are achieved, and the product quality uniformity and the production yield are remarkably improved.

Description

Technical Field

[0001] This invention belongs to the field of artificial intelligence, specifically relating to an intelligent method and system for monitoring the processing of tamarind pulp. Background Technology

[0002] With the rapid development of the deep processing industry for specialty agricultural products, tamarind, a tropical fruit rich in tartaric acid and natural sugars, has seen its pulp processing widely applied in the food, health product, and pharmaceutical raw material fields. The quality of tamarind pulp is highly dependent on the precise control of the drying process, and the drying effect is directly affected by the initial ripeness of the pulp. Currently, the industry generally adopts a "fixed parameter drying" model, where washed pulp is uniformly dehydrated at a constant temperature of 65℃, completely ignoring the natural differences between individual fruits in sugar-acid ratio, moisture content, and skin structure. This extensive process leads to a large amount of pulp charring due to overheating or mold growth due to insufficient dehydration, resulting in economic losses.

[0003] Dynamic monitoring of tamarind pulp maturity is a key prerequisite for achieving precise drying. Existing technologies attempt to roughly assess maturity by visually judging the firmness of the pulp, but this method is highly subjective, inefficient, and cannot quantify core physicochemical indicators such as the sugar-acid ratio, making it difficult to meet the requirements of modern food safety production systems such as ISO 22000.

[0004] Some automation solutions introduce single-point sensors. For example, the patent with publication number CN114526890A uses an infrared thermometer to only obtain surface temperature, which cannot reflect the internal moisture migration status. The Chinese patent with publication number CN113237876B, although deploying a visible light camera, does not solve the problem of image distortion caused by water stains after cleaning, resulting in a high false detection rate.

[0005] In summary, the existing technologies have failed to construct a closed-loop control system of "perception-decision-execution," revealing the following technical defects: First, there is a lack of non-contact quantitative indicators for ripeness measurement that integrate surface texture and internal light transmission characteristics, making it difficult for single-band imaging to comprehensively characterize the ripeness state; Second, the deep learning model has not been designed for lightweight industrial edge devices, resulting in severe real-time performance deficiencies; Third, the temperature control logic of the drying system is rigid, relying solely on environmental feedback while ignoring the dynamic impact of the fruit ripeness gradient on heat transfer efficiency.

[0006] Therefore, there is an urgent need for a method and system for monitoring tamarind pulp processing that integrates dual-band spectral sensing, lightweight intelligent reasoning, and dynamic parameter mapping, in order to achieve closed-loop control of the entire process, from non-contact quantification of ripeness to millisecond-level adaptive adjustment of drying parameters. Summary of the Invention

[0007] The technical problem to be solved by this invention is to provide an intelligent method and system for monitoring the processing of tamarind pulp, which aims to overcome the problems of pulp charring and excessive moisture residue caused by the use of fixed parameter drying mode in the prior art, and to solve the defects of strong subjectivity of manual detection, single-point sensor information dimension, and the inability of traditional machine vision models to meet the real-time and edge computing deployment requirements of industry.

[0008] To address the aforementioned technical issues, this invention provides an intelligent method for monitoring tamarind fruit-pulp processing. By constructing a dual-band spectral imaging and data deep fusion mechanism, composite feature information capable of simultaneously characterizing the surface texture and internal moisture state of the pulp is obtained. A lightweight neural network model designed for industrial edge computing environments is used to analyze this composite feature information in real time, accurately quantifying the maturity index of each piece of tamarind pulp. Finally, based on the maturity index, a dynamic temperature control model is used to generate and execute targeted drying temperature control commands, realizing a complete technical chain from high-precision online monitoring to adaptive closed-loop control, thereby significantly improving the quality uniformity and production yield of dried tamarind pulp products.

[0009] According to one aspect of the present invention, an intelligent method for monitoring tamarind pulp processing is provided, comprising the following steps:

[0010] A dual-band spectral image acquisition device is used to simultaneously acquire visible light and near-infrared transmission images of a single tamarind pulp on the conveyor belt; the dual-band spectral image acquisition device is located after the tamarind pulp washing process and before it enters the drying chamber.

[0011] The acquired visible light images are subjected to texture feature enhancement processing, and a quantized surface texture roughness index is extracted from them; the acquired near-infrared transmission images are normalized, and a quantized internal transmittance index is extracted from them.

[0012] The surface texture roughness index and the internal light transmittance index are input into a preset maturity quantification model to calculate the maturity index that characterizes the current overall maturity status of tamarind pulp.

[0013] The maturity index is input into a dynamic temperature control model, which calculates and generates a target drying temperature setpoint based on the deviation between the maturity index and the preset optimal maturity target value.

[0014] The target drying temperature setpoint is sent to the temperature control system of the drying chamber. The temperature control system drives the heating execution unit inside to precisely adjust the temperature of the chamber where the tamarind pulp is located to the target drying temperature setpoint.

[0015] In one embodiment of the present invention, the dual-band spectral image acquisition device includes a visible light imaging unit, a near-infrared imaging unit, a composite light source system, and a hardware synchronous trigger controller. The visible light imaging unit employs a monochromatic image sensor with a center wavelength of 450 nanometers and is equipped with a linear polarizing filter to eliminate reflections from water stains on the fruit pulp surface. The near-infrared imaging unit employs an image sensor with a center wavelength of 940 nanometers. The composite light source system includes a ring-shaped visible light source positioned above the conveyor belt and a planar array near-infrared backlight source positioned below the conveyor belt. The conveyor belt is made of a material transparent to 940-nanometer infrared light. The hardware synchronous trigger controller is electrically connected to a rotary encoder on the conveyor belt drive shaft. Based on the conveyor belt's movement displacement signal, it simultaneously sends acquisition trigger commands to both the visible light imaging unit and the near-infrared imaging unit, ensuring that the same tamarind pulp is imaged at the same time.

[0016] As one embodiment of the present invention, the specific steps for extracting and quantifying the surface texture roughness index include: First, using a preset two-dimensional Gabor filter bank to perform multi-directional and multi-scale convolution filtering on the visible light image to enhance the texture details related to the crystallization of sugars on the fruit pulp surface; then, applying a local binary mode algorithm to the filtered image to generate a feature histogram describing the texture structure of the pixel neighborhood; finally, calculating the statistical standard deviation of the feature histogram and using the standard deviation value as the surface texture roughness index.

[0017] As one embodiment of the present invention, the specific steps for extracting and quantifying the internal transmittance index include: first, performing target segmentation on the near-infrared transmission image to accurately identify the pixel region of tamarind pulp; then, calculating the grayscale values ​​of all pixels within the pulp pixel region to obtain an average grayscale value; finally, comparing the average grayscale value with the pre-calibrated background grayscale value of an empty conveyor belt, and calculating the internal transmittance index through a normalization function, which is inversely correlated with the internal moisture content of the pulp.

[0018] In one embodiment of the present invention, the maturity quantification model is a linear weighted summation formula: Maturity Index M = α*R + β*T, where M is the maturity index, R is the surface texture roughness index, and T is the internal light transmittance index. α is the surface texture roughness weighting coefficient, with a value of 0.65; β is the internal light transmittance weighting coefficient, with a value of 0.35. The values ​​of the weighting coefficients α and β are determined by performing dual-band imaging on a large number of tamarind pulp samples and then performing multiple linear regression analysis on the corresponding R and T values ​​with two gold standard physicochemical indicators: tartaric acid content determined by high-precision liquid chromatography and moisture content determined by Karl Fischer titration.

[0019] In one embodiment of the present invention, the maturity quantification model is a pre-trained lightweight neural network model, specifically a spatial channel compression-excited ghost network. The input layer of the network receives a two-dimensional feature vector composed of the surface texture roughness index and the internal transmittance index. The main structure of the network consists of multiple stacked ghost modules, used to effectively extract features while reducing computational complexity; and spatial and channel compression excitation modules are inserted between key feature layers to adaptively enhance feature dimensions that contribute highly to maturity determination. The output layer of the network is a fully connected layer, outputting a scalar value, namely the maturity index.

[0020] In one embodiment of the present invention, the dynamic temperature control model is a proportional-integral-derivative controller based on fuzzy logic. The controller receives the calculated maturity index as a process variable and a preset optimal maturity target value as a setpoint. The controller internally includes a fuzzy rule base, which dynamically adjusts the three control parameters (proportional, integral, and derivative) based on the deviation (E) between the maturity index and the target value, and the rate of change (EC) of the deviation. The core transfer function of the dynamic temperature control model is: Target drying temperature setpoint T_target = T_baseline - f(Kp, Ki, Kd) * (M - M_optimal), where T_baseline is a preset base drying temperature, f is the comprehensive adjustment function output by the fuzzy logic controller, M is the currently measured maturity index, and M_optimal is the optimal maturity target value.

[0021] According to another aspect of the present invention, an intelligent system for monitoring tamarind pulp processing is provided, comprising:

[0022] A dual-band spectral image acquisition module is used to simultaneously acquire visible light and near-infrared transmission images of a single tamarind pulp on a conveyor belt;

[0023] An edge computing processing module is connected to the dual-band spectral image acquisition module for performing image processing and model inference operations.

[0024] A temperature control module is signal-connected to the edge computing processing module and the heating execution unit of the drying chamber;

[0025] The edge computing processing module is configured to: perform texture feature enhancement processing on the received visible light image and extract a quantized surface texture roughness index; perform normalization processing on the received near-infrared transmission image and extract a quantized internal transmittance index; input the surface texture roughness index and the internal transmittance index into a preset maturity quantification model to calculate a maturity index; input the maturity index into a dynamic temperature control model to calculate and generate a target drying temperature setpoint, and send the setpoint to the temperature control module.

[0026] The temperature control module is configured to receive the target drying temperature setpoint and generate a corresponding pulse width modulation control signal to drive the heating execution unit in the drying chamber to adjust the power until the temperature sensor feedback value in the chamber matches the target drying temperature setpoint.

[0027] As one embodiment of the present invention, the physical configuration of the dual-band spectral image acquisition module is as follows: a gantry support fixed above the conveyor belt, the visible light imaging unit and the near-infrared imaging unit are mounted side by side on the crossbeam of the gantry support, and the lens is vertically downward.

[0028] The annular visible light source is coaxially mounted on the outer ring of the lens of the visible light imaging unit. The area array near-infrared backlight source is integrated inside the support platform below the conveyor belt. The conveyor belt is made of polytetrafluoroethylene mesh, which has a light transmittance of more than 95% in the 940-nanometer wavelength band.

[0029] The hardware synchronous trigger controller is an independent programmable logic controller. Its input is connected to an incremental rotary encoder installed on the shaft end of the conveyor belt drive roller, and its output is connected to the external trigger pins of the two imaging units respectively.

[0030] In one embodiment of the present invention, the hardware carrier of the edge computing processing module is an embedded computing device, which integrates a tensor processing unit with a computing power of not less than 0.5 trillion floating-point operations per second, and a dynamic random access memory of not less than four gigabytes. The mature quantization model and the dynamic temperature control model are stored in the non-volatile memory of the embedded computing device.

[0031] In one embodiment of the present invention, the drying chamber is divided into multiple independent temperature zones along the conveyor belt's travel direction, each zone being equipped with an independent heating execution unit and a temperature sensor. The temperature control module, based on the real-time speed of the conveyor belt and the position tracking information of the tamarind pulp, accurately applies the target drying temperature setpoint generated by the edge computing processing module to the corresponding temperature zone where the tamarind pulp is about to enter, thereby achieving personalized, segmented, dynamic temperature drying treatment for pulp at different ripeness levels on the conveyor belt.

[0032] In summary, this application includes at least one of the following beneficial technical effects:

[0033] I. This invention achieves precise quantitative monitoring of tamarind pulp maturity. By fusing surface texture information from visible light images with internal light transmission information from near-infrared images, a two-dimensional maturity evaluation system is constructed. This overcomes the shortcomings of incomplete information from a single sensor and can resist environmental interference such as surface water stains. Thus, maturity assessment is elevated from subjective qualitative judgment to objective quantitative calculation, and its quantitative accuracy is highly correlated with laboratory physicochemical test results.

[0034] Second, it ensures the high-speed real-time processing capability of industrial production lines. The lightweight neural network model or linear weighted model adopted in this invention has low computational complexity and fast inference speed. It can be deployed on low-cost industrial edge computing devices and complete the entire process from image acquisition to control command output within a processing latency of less than one second, fully meeting the real-time requirements of high-speed operation of industrial conveyor belts.

[0035] Third, an intelligent processing mode from dynamic monitoring to closed-loop adaptive control has been established. This invention directly links a precisely quantified maturity index with a dynamic temperature control model, so that the drying temperature is no longer a fixed process parameter, but a variable that is adjusted in real time according to the actual state of each tamarind pulp. This closed-loop feedback control mechanism fundamentally solves the problem of uneven drying caused by individual differences in pulp, and can control the pulp charring rate and moisture residue excess rate to extremely low levels, significantly improving the quality of the final product and production efficiency.

[0036] Fourth, it enhances the system's robustness and automation level. The entire system requires no manual intervention, automatically completing online detection and process parameter control for every product on the production line. This reduces reliance on operator experience, improves the standardization and intelligence of the production process, and provides an effective technical approach for the food processing industry to achieve refined and high-quality production. Attached Figure Description

[0037] Figure 1 This is a schematic diagram of the intelligent method for monitoring the processing of tamarind pulp according to the present invention;

[0038] Figure 2 This is a schematic diagram of the framework of the tamarind pulp processing monitoring system of the present invention. Detailed Implementation

[0039] This invention provides an intelligent method and system for monitoring tamarind pulp processing, aiming to solve the problems of charring and excessive moisture residue caused by the use of fixed parameters in existing tamarind pulp drying processes. This invention achieves precise quantification of the maturity of individual tamarind pulp pieces by constructing a dual-band spectral imaging and data deep fusion mechanism, and dynamically generates drying temperature control commands based on the quantification results, forming a closed-loop control chain from sensing, analysis to execution. The specific implementation steps of this method will be described in detail below, and the system structure supporting the operation of this method will be disclosed as necessary.

[0040] Example 1

[0041] The intelligent method for monitoring tamarind pulp processing includes the following steps: S1, using a dual-band spectral image acquisition device to simultaneously acquire visible light images and near-infrared transmission images of a single tamarind pulp on the conveyor belt;

[0042] S2, perform texture feature enhancement processing on the acquired visible light image and extract the quantized surface texture roughness index from it; perform normalization processing on the acquired near-infrared transmission image and extract the quantized internal transmittance index from it.

[0043] S3. Input the surface texture roughness index and the internal light transmittance index into a preset maturity measurement model to calculate the maturity index that represents the overall maturity status of the tamarind pulp.

[0044] S4. Input the maturity index into a dynamic temperature control model. The dynamic temperature control model calculates and generates a target drying temperature setpoint based on the deviation between the maturity index and the preset optimal maturity target value.

[0045] S5 sends the target drying temperature setpoint to the temperature control system of the drying chamber. The temperature control system drives the heating execution unit inside to precisely adjust the temperature of the chamber where the tamarind pulp is located to the target drying temperature setpoint.

[0046] Reference Figures 1 to 2 In step S1, the dual-band spectral image acquisition device is positioned after the tamarind pulp washing process and before entering the drying chamber to ensure that the acquisition object is the original pulp that has been surface-washed but has not yet been heat-treated.

[0047] The device consists of a visible light imaging unit, a near-infrared imaging unit, a composite light source system, and a hardware synchronous trigger controller. The visible light imaging unit uses a monochromatic image sensor with a center wavelength of 450 nanometers, and its lens is equipped with a linear polarizing filter to suppress specular reflection interference caused by residual water film on the fruit pulp surface, thereby preserving true surface texture information.

[0048] The near-infrared imaging unit uses an image sensor with a center wavelength of 940 nanometers. This band has strong absorption characteristics for water and can effectively penetrate the pulp tissue to reflect the internal water distribution.

[0049] The composite light source system includes a ring-shaped visible light source and a near-infrared backlight source. The ring-shaped visible light source is coaxially mounted on the outer ring of the visible light imaging unit lens, providing uniform front illumination; the near-infrared backlight source is integrated into the support platform below the conveyor belt, and the 940-nanometer infrared light emitted by it penetrates the conveyor belt to illuminate the fruit pulp, forming transmission imaging conditions.

[0050] The conveyor belt is made of polytetrafluoroethylene (PTFE) mesh, a material with a light transmittance of over 95% in the 940 nm wavelength band, ensuring a high signal-to-noise ratio for near-infrared transmission images. The hardware synchronous trigger controller is an independent programmable logic controller (PLC), whose input is connected to an incremental rotary encoder at the drive roller shaft end of the conveyor belt to acquire the conveyor belt's displacement signal in real time.

[0051] When the encoder detects a single tamarind pulp entering the imaging area, the controller simultaneously sends an external trigger signal to the visible light imaging unit and the near-infrared imaging unit to ensure that both complete image acquisition of the same pulp at the same physical moment, avoiding image misalignment caused by the movement of the conveyor belt.

[0052] In step S2, the processing of the visible light image aims to extract surface texture roughness indices related to the degree of crystallization of the sugar.

[0053] First, a pre-defined two-dimensional Gabor filter bank is used to perform multi-directional, multi-scale convolution filtering on the visible light image. This filter bank contains eight directions (spaced at 45 degrees) and three scales (with standard deviations of 1.0, 1.5, and 2.0 pixels, respectively), covering texture features from microscopic crystalline particles to macroscopic surface undulations.

[0054] After filtering, each pixel generates a 24-dimensional response vector. Next, a Local Binary Pattern Algorithm (LCA) is applied to this response vector: for each pixel, the Gabor response values ​​of its eight neighboring pixels are compared with the center pixel value to generate an 8-bit binary code. This code is then mapped to a decimal number to form a local texture code. All local texture codes for the entire image are statistically analyzed to generate a 256-dimensional histogram.

[0055] Finally, the standard deviation of the histogram is calculated as the surface texture roughness index R. The larger the standard deviation value, the more uneven the surface texture, the more significant the sugar analysis, and the higher the maturity.

[0056] The processing of near-infrared transmission images aims to extract internal transmittance indices related to internal moisture content. First, target segmentation is performed: the image is binarized using an adaptive thresholding method to separate the tamarind pulp region from the background region. During segmentation, morphological closing operations are used to eliminate tiny pores at the pulp edges, ensuring region integrity. Then, the average grayscale value I_fruit of all pixels within the pulp region is calculated. Simultaneously, the average grayscale value I_background of the conveyor belt background is pre-calibrated under no-load conditions. Finally, the internal transmittance index T is calculated using a normalization function, expressed as: The index T is inversely correlated with the internal moisture content of the fruit pulp: the higher the moisture content, the stronger the absorption of 940 nm infrared light, the lower the intensity of transmitted light, and the smaller the T value; conversely, the lower the moisture content, the larger the T value.

[0057] In step S3, the maturity quantification model is used to fuse the surface texture roughness index R and the internal transmittance index T to generate a unified maturity index M. This model can be implemented in two ways. The first is a linear weighted summation formula:

[0058]

[0059] Wherein, α is the surface texture roughness weighting coefficient, with a value of 0.65; β is the internal transmittance weighting coefficient, with a value of 0.35. These weighting coefficients were determined through offline calibration: 500 batches of tamarind pulp samples were collected, and their R and T values ​​were obtained. Simultaneously, tartaric acid content was determined using high-precision liquid chromatography, and moisture content was determined using Karl Fischer titration. The tartaric acid and moisture contents were weighted and normalized to construct a gold standard label for maturity. Through multiple linear regression analysis, the combination of α and β that minimizes the prediction error was solved, ultimately determining the aforementioned weighting values. This linear model has extremely low computational complexity and is suitable for scenarios with extremely high real-time requirements.

[0060] The second implementation is a lightweight neural network model, specifically a spatial channel compression-excited ghost network. The input layer of this network receives a two-dimensional feature vector [R, T]. The main body of the network consists of four stacked ghost modules. Each ghost module first generates a partial feature map through a regular convolutional layer, and then generates the remaining feature map through inexpensive linear operations (such as depthwise separable convolution), thereby significantly reducing the number of parameters while maintaining expressive power.

[0061] Following the second and third ghost modules of the network, a spatial and channel compression activation module is inserted. This module first performs global average pooling on the feature map to generate channel description vectors; then, it calculates the attention weights for each channel using two fully connected layers and a sigmoid activation function; finally, it multiplies the weights with the original feature map to enhance the channels that contribute significantly to maturity discrimination. The network output layer is a fully connected layer that outputs a single scalar value M. The model's inference time on the Jetson Nano embedded platform is 320 milliseconds, meeting the real-time processing requirements at a conveyor belt speed of 0.5 meters per second (single fruit pulp imaging region width of 150 millimeters, processing window of 300 milliseconds).

[0062] In step S4, the dynamic temperature control model receives the maturity index M as a process variable and the preset optimal maturity target value M_optimal as the setpoint. This model employs a proportional-integral-derivative controller based on fuzzy logic. The controller has a pre-set fuzzy rule base, and its inputs are the deviation E = M_optimal - M and the deviation change rate EC = E(k) - E(k-1). Both E and EC are divided into seven fuzzy sets: negative large, negative medium, negative small, zero, positive small, positive medium, and positive large.

[0063] The output consists of adjustments to the proportional coefficient Kp, integral coefficient Ki, and differential coefficient Kd, also divided into seven fuzzy sets. There are forty-nine fuzzy rules, such as: "If E is positive and EC is positive, then Kp is negative, Ki is zero, and Kd is positive," indicating that the current maturity is far below the target and still deteriorating, requiring a significant reduction in drying temperature to avoid coking, while simultaneously enhancing the differential action to suppress overshoot. Fuzzy inference uses the Mamdani method, and defuzzification uses the centroid method. Finally, the target drying temperature setpoint T_target is calculated using the following formula:

[0064]

[0065] Wherein, T_reference is the preset baseline drying temperature, usually set to 65 degrees Celsius; f is the comprehensive adjustment function output by the fuzzy controller, with a value range of 0.8 to 1.2, ensuring that the temperature adjustment range is within a reasonable range. This model can dynamically adjust the control intensity according to the magnitude and trend of maturity deviation, avoiding the overshoot or slow response problems caused by traditional PID controllers when parameters are fixed.

[0066] In step S5, the target drying temperature setpoint is sent to the temperature control system of the drying chamber. After receiving the setpoint, the system generates a corresponding pulse width modulation signal to drive the heating actuator (such as a ceramic heating tube or a hot air circulating fan) to adjust its power.

[0067] The drying chamber incorporates a high-precision platinum resistance temperature sensor, providing real-time feedback of the actual temperature. The control system employs a closed-loop feedback mechanism, continuously comparing the feedback value with the set value until the error is less than 0.5 degrees Celsius. Furthermore, the drying chamber is divided into six independent temperature zones along the conveyor belt's travel direction, each equipped with an independent heating unit and temperature sensor. An edge computing processing module accurately tracks the position of each tamarind pulp based on the real-time conveyor belt speed and rotary encoder counts. When the system determines that a piece of pulp is about to enter the third temperature zone, it pre-calculates the target T value for that pulp and sends it to the temperature controller of that zone, achieving segmented dynamic temperature drying. This mechanism ensures that pulp at different stages of ripeness completes drying under its optimal temperature curve, avoiding uneven quality caused by a uniform global temperature.

[0068] On the other hand, the intelligent system for monitoring tamarind pulp processing disclosed in this application specifically includes a dual-band spectral image acquisition module, an edge computing processing module, and a temperature control module.

[0069] The physical structure of the dual-band spectral image acquisition module is as described above, including a gantry support, imaging unit, light source, and synchronization controller. The edge computing processing module is carried by an embedded computing device, which integrates a tensor processing unit with a computing power of no less than 0.5 trillion floating-point operations per second and is equipped with 4 gigabytes of dynamic random access memory.

[0070] The mature quantification model and dynamic temperature control model are stored in the device's non-volatile memory as firmware, and are loaded and run upon power-on. The temperature control module consists of multiple temperature zone controllers. Each controller receives temperature setting commands from the edge computing processing module and drives the heating execution unit through pulse width modulation, while simultaneously collecting feedback from temperature sensors to form a local closed loop.

[0071] The system's workflow is as follows: After being washed, tamarind pulp enters a conveyor belt. While passing through a dual-band spectral image acquisition module, visible light and near-infrared images are simultaneously acquired. The image data is transmitted to an edge computing processing module via gigabit Ethernet. The module executes steps S2 to S4 to calculate the target drying temperature. Based on the pulp's location information, this temperature value is sent to the temperature control module of the corresponding temperature zone. The temperature control module adjusts the heating power to stabilize the temperature in that zone at the set value. The pulp then enters that temperature zone to complete the drying process. The entire process requires no manual intervention, with a processing delay of less than 800 milliseconds, meeting the requirements of industrial production cycle time.

[0072] Through the above methods and systems, this invention achieves objective quantification, high-speed real-time processing, and closed-loop adaptive drying control of tamarind pulp maturity. Actual test data shows that after adopting this solution, the pulp charring rate decreased from 32.7% to 2.1%, the moisture residue exceeding the standard rate decreased from 28.5% to 1.8%, and the tartaric acid pyrolysis rate was controlled within 5%, significantly improving product quality and economic benefits.

[0073] 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 present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention. Therefore, the embodiments should be regarded as exemplary and non-limiting in all respects.

[0074] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. An intelligent method for monitoring tamarind pulp processing, characterized in that, include: Visible light and near-infrared transmission images of individual tamarind pulps on the conveyor belt are acquired simultaneously using a dual-band spectral image acquisition device, which is positioned after the tamarind pulp washing process and before it enters the drying chamber. The visible light image is subjected to texture feature enhancement processing, and a quantized surface texture roughness index is extracted from it; the near-infrared transmission image is normalized, and a quantized internal transmittance index is extracted from it. The surface texture roughness index and the internal light transmittance index are input into the preset maturity measurement model to calculate the maturity index, which represents the overall maturity status of the tamarind pulp. The maturity index is input into the dynamic temperature control model, which calculates and generates the target drying temperature setpoint based on the deviation between the maturity index and the preset optimal maturity target value. The target drying temperature setpoint is sent to the temperature control system of the drying chamber. The temperature control system drives the heating execution unit inside to precisely adjust the temperature of the chamber where the tamarind pulp is located to the target drying temperature setpoint.

2. The intelligent method for monitoring tamarind pulp processing according to claim 1, characterized in that, The dual-band spectral image acquisition device includes a visible light imaging unit, a near-infrared imaging unit, a composite light source system, and a hardware synchronous trigger controller; The visible light imaging unit uses a monochrome image sensor with a center wavelength of 450 nanometers and is equipped with a linear polarizing mirror; The near-infrared imaging unit uses an image sensor with a center wavelength of 940 nanometers; The composite light source system includes a ring-shaped visible light source positioned above the conveyor belt and an array-type near-infrared backlight source positioned below the conveyor belt. The conveyor belt is made of a material that is transparent to infrared light in the 940-nanometer band. The hardware synchronous trigger controller is electrically connected to the rotary encoder on the conveyor belt drive shaft and is used to synchronously trigger the visible light imaging unit and the near-infrared imaging unit according to the conveyor belt travel displacement signal.

3. The intelligent method for monitoring tamarind pulp processing according to claim 1, characterized in that, The extracted surface texture roughness indicators include: A pre-set two-dimensional Gabor filter bank was used to perform multi-directional and multi-scale convolution filtering on the visible light image to enhance the texture details related to crystallization of sugars on the fruit pulp surface. The local binary mode algorithm is applied to the filtered image to generate feature histograms describing the texture structure of the pixel neighborhood; the statistical standard deviation of the feature histograms is calculated, and the standard deviation is used as an index of surface texture roughness.

4. The intelligent method for monitoring tamarind pulp processing according to claim 1, characterized in that, The internal transmittance index includes: Target segmentation is performed on near-infrared transmission images to identify pixel regions of tamarind pulp; Calculate the average grayscale value of all pixels within the fruit pulp pixel region; The internal light transmittance index is calculated by normalizing the average gray value and the pre-calibrated background gray value of the empty conveyor belt using a normalization function. This index is inversely correlated with the moisture content inside the fruit pulp.

5. The intelligent method for monitoring tamarind pulp processing according to claim 1, characterized in that, The maturity quantification model is a pre-trained lightweight neural network model, specifically a spatial channel compression excitation ghost network. The network's input layer receives a two-dimensional feature vector consisting of a surface texture roughness index and an internal transmittance index. The main structure of the network consists of multiple ghost modules stacked together, with spatial and channel compression excitation modules inserted between key feature layers. The network's output layer is a fully connected layer, and the output scalar value serves as the maturity index.

6. The intelligent method for monitoring tamarind pulp processing according to claim 1, characterized in that, The dynamic temperature control model is a proportional-integral-derivative controller based on fuzzy logic; The controller receives the maturity index as a process variable and the preset optimal maturity target value as a setpoint. The controller contains a fuzzy rule base and dynamically adjusts the proportional, integral, and derivative control parameters based on the deviation between the maturity index and the target value and its rate of change.

7. The intelligent method for monitoring tamarind pulp processing according to claim 1, characterized in that, The drying chamber is divided into multiple independent temperature zones along the direction of the conveyor belt, and each temperature zone is equipped with an independent heating actuator and temperature sensor. The temperature control system precisely applies the target drying temperature setpoint to the corresponding temperature zone that the tamarind pulp is about to enter, based on the real-time speed of the conveyor belt and the position tracking information.

8. An intelligent system for monitoring tamarind pulp processing, characterized in that, include: A dual-band spectral image acquisition module is used to simultaneously acquire visible light and near-infrared transmission images of a single tamarind pulp on the conveyor belt; The edge computing processing module is connected to the dual-band spectral image acquisition module. It is used to perform texture feature enhancement processing on the received visible light image to extract the surface texture roughness index, and to perform normalization processing on the received near-infrared transmission image to extract the internal transmittance index. The surface texture roughness index and the internal transmittance index are input into the preset maturity quantification model to calculate the maturity index, and the maturity index is input into the dynamic temperature control model to generate the target drying temperature setpoint. The temperature control module is connected to the edge computing processing module and the heating execution unit of the drying chamber. It is used to receive the target drying temperature setpoint and generate a pulse width modulation control signal to drive the heating execution unit to adjust the power until the feedback value of the temperature sensor in the chamber matches the target drying temperature setpoint.

Images