Data acquisition and analysis system for grain processing, processing device and processing method

Abstract

The invention provides a data acquisition and analysis system for grain processing, a processing device and a processing method, and relates to the technical field of grain processing. The data acquisition and analysis system comprises an image acquisition module, a light source module, a component sensing module, a specification adaptation module, a data processing module and a man-machine interaction module; the appearance details of processed products are captured through high-definition images, internal components are analyzed through spectral analysis, temperature, humidity and dust are monitored through environmental sensing, dominant and hidden defects such as cracks and excessive moisture can be recognized at the same time, the detection effect is improved, different products can be automatically recognized, corresponding detection standards can be called, the parameter debugging time during production changing is effectively shortened, and the production efficiency is improved. According to the full-process automatic processing device, the requirements for multi-variety and small-batch production are met, in addition, through the full-process automatic processing device, manual intervention is greatly reduced, the labor cost of a production line is effectively reduced, the counting precision is improved, quantitative packaging errors are avoided, and therefore the intelligent level and the production efficiency of grain processing are effectively improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of grain processing, and in particular to a data acquisition and analysis system for grain processing, a processing device and a processing method. BACKGROUND

[0002] In the current grain processing production line, counting and packing and good product identification are key links to ensure product quality and production efficiency. Traditional processing methods rely on manual counting and visual screening, which not only have high labor intensity and low efficiency, but also are prone to cause counting errors and good product misjudgment due to human factors.

[0003] With the development of automation technology, some production lines have introduced simple photoelectric counting devices and image recognition equipment, but these devices generally have the problems of single function and poor adaptability. For example, it is difficult to fully capture the appearance defects of the processed products through single-angle image acquisition, the analysis of the internal composition of the processed products is lacking, which leads to one-sided good product determination standards, and the detection parameters cannot be automatically adjusted according to different specifications of the processed products, frequent manual intervention is required, which seriously affects the production continuity and product quality stability. SUMMARY

[0004] The present application provides a data acquisition and analysis system for grain processing, which can realize multi-dimensional data acquisition, comprehensive analysis and intelligent judgment of grain processing products, and has good specification adaptability and human-computer interaction capability, solves the problems of low counting accuracy, incomplete good product identification and poor adaptability in the existing grain processing process, and improves the intelligent level and production efficiency of grain processing through the integration of multi-dimensional data acquisition, intelligent analysis and automatic control functions.

[0005] Therefore, the first aspect of the present application provides a data acquisition and analysis system for grain processing, comprising an image acquisition module comprising at least two high-definition cameras and a spectral imaging unit arranged at different angles, the high-definition cameras being used to acquire appearance images of the processed products, and the spectral imaging unit being used to acquire component distribution spectrum data of the processed products; a light source module arranged in cooperation with the image acquisition module, comprising switchable visible light sources and near-infrared light sources, for providing adaptive lighting for different types of data acquisition; a component sensing module comprising a temperature and humidity sensor and a weight sensor, for acquiring processing environment parameters; a specification adaptation module, which internally stores parameter models of multiple different specifications of processed products, and can automatically match corresponding detection parameters, including image acquisition accuracy, spectral analysis range and counting determination standards, by identifying the initial characteristics of the processed products; A data processing module is connected with the image acquisition module, the light source module and the composition perception module respectively, and is configured to perform feature fusion on the appearance and morphology image and the composition distribution spectrum data, combine the processing environment parameters to construct a comprehensive evaluation dimension containing physical features and composition features, and determine whether the processing is a good product and count the good products. A man-machine interaction module includes a touch display screen and a voice interaction unit, and is configured to manually input parameters, query detection data and receive voice prompt information of the system by an operator, and support temporary adjustment of the operation mode of the system through external instructions.

[0006] In some embodiments, the data processing module includes: A deep learning fusion module is configured to perform feature fusion on the appearance and morphology image and the composition distribution spectrum data, combine the processing environment parameters to construct a comprehensive evaluation dimension containing physical features and composition features. A dynamic threshold generation module is configured to automatically adjust the good product determination threshold according to the batch characteristics of the processed products in real time after obtaining the detection parameters of the specification adaptation module. A counting compensation algorithm module is configured to eliminate the counting error of the processed products in the overlapping and inclined state.

[0007] In some embodiments, the deep learning fusion module adopts a double-branch network structure of an attention mechanism, and includes A first branch is configured to extract the shape distortion, surface crack and color uniformity features in the appearance and morphology image. A second branch is configured to analyze the moisture content, oil distribution and additive concentration features in the composition distribution spectrum data. The attention mechanism is configured to automatically allocate the feature weights of the two branches according to the type of the processed products.

[0008] The data acquisition and analysis system according to claim 2, wherein the dynamic threshold generation module includes A data feature extraction unit is configured to extract the key feature data such as the raw material ratio, the processing temperature curve and the molding pressure from the batch of processed products collected in real time. A threshold calculation unit is configured to establish a threshold calculation model by analyzing the historical qualified batch data based on a machine learning algorithm, and calculate the good product determination threshold of the current batch in combination with the real-time feature data. A threshold updating unit is configured to dynamically correct the calculated threshold according to the environmental parameter fluctuation, the equipment operation state change and other factors in the production process.

[0009] In some embodiments, the counting compensation algorithm module includes: A three-dimensional reconstruction subunit is configured to generate a three-dimensional profile of the processed products by using multi-angle images. The overlapping determination subunit determines the shielding relationship between the processed products by calculating the spatial coordinates of the three-dimensional profile. The morphological correction subunit corrects the projected area of the inclined processed product to achieve accurate counting.

[0010] In some embodiments, the data processing module is further configured to statistically analyze the number of defective products within a preset time period, and issue a warning signal through the human-computer interaction module when the proportion of defective products exceeds a preset threshold.

[0011] The second aspect of the application provides a grain processing device, comprising: The conveying mechanism is a belt conveyor, and the belt surface of the belt conveyor is provided with anti-skid lines; The image acquisition module, the light source module, and the component sensing module are arranged on the conveying path of the conveying mechanism. The positioning sensor is arranged on the conveying mechanism and connected with the data acquisition and analysis system, and is used to position the processed product and trigger the data acquisition and analysis system to work. The rejection mechanism is arranged on the conveying mechanism and connected with the data acquisition and analysis system, and is used to reject defective processed products from the conveying mechanism. The packaging mechanism is connected with the data acquisition and analysis system, and comprises a rack for placing packaging materials, a packaging forming assembly, and a sealing assembly, so as to quantitatively package the processed products of good products according to the counting result.

[0012] In some embodiments, the rejection mechanism comprises pneumatic push rods and a defective product collection box, which are arranged on both sides of the conveying mechanism respectively, and are used to push the processed products of defective products into the defective product collection box.

[0013] The third aspect of the application provides a grain processing method, which is applied to the above-mentioned grain processing device, and is used for counting and packaging processed products and identifying good products, and comprises the following steps: S1: conveying the processed products to be processed at a set speed through the conveying mechanism; S2: when the positioning sensor detects that the processed product reaches the image acquisition area, triggering the image acquisition module, the light source module, and the component sensing module to work, acquiring a plurality of information data of the processed product, and transmitting the plurality of information data to the data processing module; S3: the data processing module analyzes the plurality of information data, determines whether the processed product is a good product, simultaneously counts the good products by using a counting algorithm, and sends the counting result and the good product identification result to the packaging mechanism and the rejection mechanism respectively; S4: If identified as defective, the rejection mechanism rejects the processed product when it reaches the rejection position; if identified as good, the conveying mechanism conveys it to the packaging mechanism; S5: The packaging mechanism performs a packaging operation on the corresponding number of good processed products using the package forming assembly and the sealing assembly according to the received counting result when the count reaches the set number.

[0014] In some embodiments, the processing method further comprises S6: When it is detected that the packaging material is insufficient, a signal is sent to the data processing module to control the conveying mechanism to pause conveying, and a warning signal is issued through the human-computer interaction module.

[0015] The data acquisition and analysis system for grain processing, processing device and processing method provided by the present application realize technical breakthroughs in the following ways: the data acquisition and analysis system realizes integrated fusion of omnidirectional detection, automatic adaptation and efficient process control, realizes accurate quality control through multi-dimensional data acquisition, captures appearance details through high-definition images, analyzes internal components through spectral analysis, monitors temperature, humidity and dust through environmental sensing, compared with traditional single detection methods, can simultaneously identify dominant and recessive defects such as cracks and excessive moisture, and effectively improves detection coverage and accuracy; secondly, the specification adaptation module has multiple types of parameter models built-in, which can automatically identify different products and call corresponding detection standards, and cooperate with dynamic threshold generation technology to adapt to batch differences in real time, effectively reducing parameter debugging time during changeover, meeting the needs of multi-variety small-batch production; and the present application greatly reduces manual intervention through the whole process, effectively reduces labor costs of the production line, improves counting accuracy, avoids quantitative packaging errors, and thus effectively improves the intelligent level and production efficiency of grain processing. BRIEF DESCRIPTION OF DRAWINGS

[0016] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiments. It should be understood that the following drawings only show some embodiments of the present application, and therefore should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can also be obtained without creative labor on the basis of these drawings.

[0017] Figure 1 The module architecture diagram of the data acquisition and analysis system provided by the embodiment of the present application; Figure 2 The module architecture diagram of the data processing module provided by the embodiment of the present application; Figure 3 The module architecture diagram of the dynamic threshold generation module provided by the embodiment of the present application; Figure 4 The module architecture diagram of the calculation compensation algorithm module provided by the embodiment of the present application; Figure 5 A schematic diagram of a grain processing device structure is provided for the embodiments of the present application. Figure 6 A flow chart of a grain processing method is provided for the embodiments of the present application.

[0018] Reference signs: 100, conveying mechanism; 200, data acquisition and analysis system; 300, positioning sensor; 400, rejection mechanism; 410, pneumatic push rod; 420, defective product collection box; 500, packaging mechanism. DETAILED DESCRIPTION

[0019] In order to make the objectives, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, rather than all the embodiments of the present application. The components of the embodiments of the present application described and shown in the drawings herein can be arranged and designed in various different configurations.

[0020] Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the claimed present application, but only represents selected embodiments of the present application. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present application without making creative efforts fall within the scope of protection of the present application.

[0021] It should be noted that, in this document, relational terms such as first and second and the like can be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Moreover, the terms "comprises", "comprising", or any other variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without more limitations, an element preceded by "comprises... a" does not, without more constraints, foreclose the existence of additional identical elements in the process, method, article, or apparatus that comprises the recited element.

[0022] In the embodiments of the present application, it should be understood that the disclosed apparatus and method can also be implemented in other manners. The embodiments described above are merely exemplary. For example, the flowcharts and block diagrams in the accompanying drawings illustrate the architectural, functional and operational logic of the embodiments of the present application. In this regard, each block in the flowcharts or block diagrams can represent a module, a segment or a portion of code which includes one or more executable instructions for implementing the specified logic function. It should also be noted that in some alternative implementations, the functions noted in the blocks can occur out of the order noted in the accompanying drawings. For example, two blocks noted in succession can actually be executed substantially concurrently or in reverse order, depending on the functionality involved. It should also be noted that each block in the block diagrams and / or flowcharts and combinations thereof can be implemented by a dedicated hardware-based system which performs specified functions or actions, or can be implemented by a combination of dedicated hardware and computer instructions.

[0023] In addition, each functional module in the various embodiments of the present application can be integrated together to form a separate part, or each module can exist independently, or two or more modules can be integrated to form a separate part.

[0024] If the functions are implemented in the form of software function modules and sold or used as independent products, they can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present application can be embodied in the form of a software product, and the computer software product is stored in a storage medium, and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various program codes that can be stored in the medium.

[0025] Some embodiments of the present application will be described in detail below with reference to the accompanying drawings. In the case of no conflict, each of the following embodiments and each feature in the embodiments can be combined with each other.

[0026] In the current grain biscuit processing production line, counting and packing and good product identification are the key links to ensure product quality and production efficiency. The traditional processing method relies on manual counting and visual screening, which not only has high labor intensity and low efficiency, but also is prone to counting errors and good product misjudgment due to human factors. For example, when counting manually, workers have difficulty in concentrating for a long time due to the repetitive and monotonous counting action; and in the good product identification link, the recognition ability of workers for the fine cracks and color difference on the surface of the biscuits is limited, which directly affects the product quality and enterprise reputation.

[0027] With the development of automation technology, some production lines have introduced simple photoelectric counting devices and image recognition equipment, but these devices generally have the problems of single function and poor adaptability. Taking the image recognition equipment as an example, it usually adopts a fixedly installed single camera, and it is difficult to fully capture the appearance defects of the biscuits through single-angle image acquisition. For biscuits with irregular shapes or three-dimensional structures, cracks, foreign matter attachment and other problems in recessed parts and bottoms are easily ignored. In addition, the existing equipment lacks the ability to analyze the internal composition of the biscuits, and only relies on appearance judgment to determine good products, resulting in one-sided judgment standard. In actual production, although some biscuits have good appearance, the internal moisture content exceeds the standard due to uneven baking temperature, which is prone to mold and deterioration during storage, but cannot be identified by the existing equipment. More importantly, these devices cannot automatically adjust the detection parameters according to different specifications of biscuits, and when the production line switches to produce biscuits of different sizes, thicknesses and shapes, the image resolution, detection threshold and other parameters need to be manually modified by the operator, and the production line needs to be suspended during the adjustment process, which seriously affects the production continuity and product quality stability.

[0028] To solve the above technical problems, the present application provides a data acquisition and analysis system for grain processing, a processing device and a processing method, which can realize multi-dimensional data acquisition, comprehensive analysis and intelligent judgment of grain biscuits, and has good specification adaptability and human-computer interaction capability, which can effectively improve the counting accuracy and the accuracy of good product identification, reduce labor cost and improve production efficiency.

[0029] Embodiment one The first aspect of the present application provides a data acquisition and analysis system for grain processing, which comprises Figure 1 , including: an image acquisition module, a light source module, a component sensing module, a specification adaptation module, a data processing module and a human-computer interaction module; The image acquisition module comprises at least two high-definition cameras and a spectral imaging unit arranged at different angles, the high-definition cameras are used to acquire the appearance image of the processed product, and the spectral imaging unit is used to acquire the component distribution spectrum data of the processed product; The module comprises at least two high-definition cameras and a spectral imaging unit arranged at different angles, forming a multi-dimensional and three-dimensional data acquisition network.

[0030] The high-definition camera adopts an industrial-grade high-definition imaging device with a pixel of no less than 12 million, ensuring the capture of subtle features on the surface of the processed product. Two cameras are arranged at horizontal and 45-degree angles: the horizontal camera is mainly used to shoot the top plane of the processed product, clearly recording whether there are cracks, spots, uneven color, and other appearance defects on the surface; the 45-degree oblique camera focuses on capturing the side shape of the processed product, effectively detecting whether the product has shape distortion, irregular edges, and other problems. For example, when detecting biscuits, the horizontal camera can find subtle cracks on the surface of the biscuits, and the oblique camera can identify whether the edges of the biscuits are excessively curved or incomplete.

[0031] The spectral imaging unit obtains the distribution information of the internal components of the processed product by analyzing the absorption and reflection characteristics of different wavelengths of light. This unit covers the visible light to near-infrared band, which can detect the moisture content, oil distribution, protein content, and additive distribution of the processed product. In practical applications, spectral imaging can quickly identify whether the moisture distribution in the bread is uniform, avoiding the risk of mold caused by excessive moisture or the problem of hard texture caused by insufficient moisture.

[0032] To ensure the stability of image acquisition, the camera and spectral imaging unit are installed on shockproof brackets to reduce image blurring caused by equipment vibration. At the same time, the lens is designed with dustproof to avoid the attachment of dust in the processing environment affecting the imaging quality, ensuring long-term stable operation. In this application, a near-infrared spectrometer can be selected as the spectral signal analysis unit to improve the detection sensitivity of organic components, thus more accurately obtaining component data.

[0033] The light source module is cooperatively arranged with the image acquisition module and includes switchable visible light sources and near-infrared light sources, which are used to provide adaptive lighting for different types of data acquisition. The light source module is installed around the image acquisition module, arranged in a ring or symmetrically on both sides of the camera and spectral imaging unit, providing stable light for image acquisition, avoiding image blurring and distortion caused by insufficient or uneven light.

[0034] The light source module, as a supporting facility of the image acquisition module, directly affects the quality of image and spectral data. The module includes switchable visible light sources and near-infrared light sources, which are integrated in the same light source housing and switched by the control system, providing adaptive lighting conditions according to different data acquisition needs.

[0035] In the present application, the visible light source can adopt a high-brightness LED area light source with a color temperature close to natural light (about 5500K), which can truly restore the color characteristics of the processed products. Such light source has high light uniformity, ensuring the uniform brightness of each part of the surface of the processed products, and avoiding misjudgment caused by uneven lighting. For example, when detecting cakes processed from grains, uniform visible light illumination can clearly present the color difference on the surface of the cakes, helping to identify products that are over-baked or under-baked.

[0036] The near-infrared light source adopts an LED array with specific wavelengths, mainly working with the spectral imaging unit. Near-infrared light can penetrate the surface layer of the processed products and interact with the internal molecules, thereby reflecting the information of the internal components. By switching different wavelength combinations of the near-infrared light source, different detection components can be optimized, improving the accuracy of component analysis.

[0037] The switching of the light source module is automatically controlled by the system, and is switched to the corresponding light source in real time according to the type of data acquisition (appearance image or component spectrum). In some embodiments, the data acquisition and analysis system further includes an adaptive control module connected with the data processing module to automatically adjust the exposure parameters of the image acquisition module and the spectral ratio of the light source module according to changes in environmental parameters. The exposure parameter adjustment is mainly for the high-definition camera of the image acquisition module, which adjusts the exposure time and aperture size according to the change in environmental light intensity, ensuring that the brightness of the collected image is moderate and the details are clear.

[0038] The intensity of environmental light changes at different times and under different working conditions, such as day and night, sunny and cloudy days, and the switching of lights in the workshop. When the light is strong, if the exposure parameters remain unchanged, the collected image will be too bright, resulting in the loss of details on the surface of the product; when the light is weak, the image will be too dark and cannot clearly present the product features. The adaptive control module monitors the intensity of environmental light in real time through a light sensor, and automatically adjusts the exposure time and aperture of the camera when the light intensity changes. When the light is strong, shorten the exposure time or reduce the aperture to reduce the amount of light entering the camera; when the light is weak, lengthen the exposure time or increase the aperture to increase the amount of light entering. For example, on a sunny day, the light in the workshop is sufficient, and the module will shorten the exposure time from the default 50ms to 20ms, while reducing the aperture to avoid overexposure of the image; in the evening, the light becomes dim, and the module will lengthen the exposure time to 80ms and increase the aperture to ensure appropriate image brightness.

[0039] Spectral ratio adjustment adjusts the intensity ratio of the visible and near-infrared light sources in the light source module based on environmental parameters and product type, optimizing the spectral imaging unit's acquisition performance. Changes in environmental parameters such as temperature and humidity can affect the propagation of light through the air, thereby affecting the accuracy of spectral data. For example, high humidity increases the attenuation of near-infrared light, necessitating an increase in the intensity of the near-infrared light source to ensure the quality of the spectral signal.

[0040] Different product types reflect and absorb light of different wavelengths differently, requiring different spectral profiles. For example, when inspecting light-colored biscuits, where visible light is more strongly reflected, the visible light intensity can be appropriately reduced to avoid overly bright images. However, when inspecting dark-colored cereal bars, the visible light intensity needs to be increased to clearly reveal their appearance, and the intensity of the near-infrared light source needs to be adjusted accordingly to obtain accurate ingredient information.

[0041] The adaptive control module automatically adjusts the intensity ratio of visible light and near-infrared light sources based on the environmental parameters provided by the ingredient perception module and the product type determined by the specification adaptation module. During the adjustment process, the total light intensity is kept relatively stable to avoid excessive impact on the exposure parameters of the image acquisition module. For example, when inspecting dark cereal bars, the module adjusts the intensity ratio of visible light to near-infrared light from the default 1:1 to 2:1, ensuring that the quality of both the appearance image and the spectral data meet the inspection requirements. The component sensing module includes temperature, humidity, and weight sensors to collect processing environment parameters. The component sensing module is responsible for collecting various processing environment parameters and providing environmental background information for quality analysis of processed products. This module includes temperature, humidity, and weight sensors to comprehensively monitor key environmental factors during the processing process.

[0042] Temperature and humidity sensors are installed in various locations throughout the processing workshop, including around conveyor belts, near processing equipment, and in corners. They monitor changes in ambient temperature and relative humidity in real time. Temperature and humidity have a significant impact on the quality of processed grain products. For example, excessively high temperatures cause products like biscuits to lose moisture quickly, making them brittle and fragile; while excessively high humidity can dampen products like bread, affecting their taste and shelf life. By monitoring temperature and humidity in real time, the system provides a reference for subsequent quality assessments. If temperature and humidity exceed normal ranges, the system can prompt operators to adjust the workshop environment.

[0043] Weight sensors are installed at specific locations on the conveyor mechanism to accurately measure the weight of processed products as they pass through. Weight is a key indicator of product conformance to specifications. For example, each standard biscuit has a specific weight requirement. Too light indicates insufficient raw materials, while too heavy can affect packaging and transportation. The high measurement accuracy of weight sensors can detect even the smallest weight discrepancies, ensuring that substandard products are promptly screened out.

[0044] At the same time, each sensor collects data at a set frequency and transmits it to the data processing module through wired or wireless means, providing relevant parameters of the processing environment to the data processing module as a reference basis for good product determination.

[0045] The specification adaptation module has multiple parameter models for different specifications of processed products. It can automatically match the corresponding detection parameters, including image acquisition accuracy, spectral analysis range, and counting determination criteria, by recognizing the initial characteristics of the processed products.

[0046] The specification adaptation module is the core component of the system for detecting different types and specifications of processed products. It has multiple parameter models for common grain processed products built-in, which can automatically match the corresponding detection parameters based on the initial characteristics of the processed products.

[0047] The parameter model library of this unit covers multiple common grain processed products such as biscuits, bread, cereal bars, etc. Each product contains multiple different specifications. For example, the biscuit model library includes different types such as grain biscuits, cookie biscuits, and sandwich biscuits, each with different size, thickness, and other specification parameters. These parameter models are established based on a large amount of experimental data and production experience, containing information such as appearance feature standards, ingredient content range, and size tolerance of various products.

[0048] When the processed product enters the detection area, the specification adaptation module first obtains the initial characteristics of the product such as shape, size, color, etc. through the image acquisition module. Then, these characteristics are compared with the built-in parameter models to quickly identify the type and specification of the product. During the identification process, a feature matching algorithm is used to extract key feature points of the product and compare them with the features in the model library to ensure accuracy and efficiency of the matching.

[0049] After matching, the specification adaptation module automatically calls the detection parameters corresponding to the type of product, including image acquisition accuracy, spectral analysis range, and counting determination criteria. For example, for a relatively flat soda biscuit, the image acquisition accuracy can be appropriately reduced; while for a cookie biscuit with complex patterns on the surface, the image acquisition accuracy needs to be improved to capture subtle pattern features. The spectral analysis range will also be adjusted according to the ingredient characteristics of the product, such as detecting sandwich biscuits containing more oil, which will focus on analyzing the spectral band related to oil.

[0050] The data processing module is connected with the image acquisition module, light source module, and ingredient sensing module, respectively, for feature fusion of appearance image and ingredient distribution spectrum data, combined with processing environment parameters to construct a comprehensive evaluation dimension containing physical and ingredient characteristics, to determine whether the processing is good and count the good products. The data processing module is responsible for comprehensive processing and analysis of various types of data collected by the image acquisition module, light source module, and composition perception module, ultimately determining whether the processed product is a good product and completing the counting.

[0051] This module can use high-performance embedded processing chips with powerful data processing and processing capabilities, capable of quickly processing large amounts of image and spectral data. The processing process mainly includes data preprocessing, feature extraction, feature fusion, and judgment counting.

[0052] In the data preprocessing stage, the collected images are processed for noise reduction, enhancement, etc., to remove interference information in the images and highlight product features; spectral data are smoothed to eliminate noise effects. In the feature extraction stage, the appearance features of the product, such as shape, color, texture, and defects, are extracted from the preprocessed images; the composition features of the product, such as moisture content, oil content, and additive concentration, are extracted from the spectral data.

[0053] The feature fusion link comprehensively analyzes the appearance features and composition features, combines the environmental parameters collected by the composition perception module, and constructs a comprehensive evaluation dimension containing physical and composition features. For example, for a biscuit, not only whether its surface has cracks (physical feature) is considered, but also whether its moisture content is within a reasonable range (composition feature) is considered, and the influence of environmental temperature and humidity on product quality is considered for comprehensive evaluation.

[0054] In the judgment process, the comprehensive evaluation results are compared with the judgment criteria provided by the specification adaptation module. If it meets the criteria, it is determined as a good product, otherwise it is determined as a defective product. The counting function counts the number of products determined as good products. The statistical process considers the arrangement state of the products, such as whether there is overlap, tilt, etc., and corrects it through the corresponding algorithm to ensure the accuracy of the count.

[0055] The human-computer interaction module includes a touch display screen and a voice interaction unit, which is used for manual input of parameters by the operator, query of detection data, and reception of voice prompt information from the system, and supports temporary adjustment of the operation mode of the system through external instructions.

[0056] The human-computer interaction module is a component for information exchange between the operator and the system, which facilitates the operator to set, monitor and operate the system, and includes a touch display screen and a voice interaction unit.

[0057] In this embodiment, the touchscreen uses a capacitive screen, allowing operators to set various parameters by touching the screen, such as selecting product specifications, adjusting detection accuracy, and setting warning thresholds. The screen displays real-time system operating status, including current detection speed, yield rate, and the number and type of defective products, allowing operators to keep abreast of production progress. The screen also allows for querying historical detection data, such as quality statistics for different batches of products and equipment operation records, providing data support for production management and quality analysis.

[0058] The voice interaction unit supports voice commands and voice prompts. Operators can use voice commands to operate the system, such as "start testing," "pause operation," and "check yield rate," improving operational convenience. The system also provides voice prompts to operators, such as "High defective rate detected" and "Insufficient packaging materials," allowing operators to obtain timely system information while focusing on other tasks.

[0059] Furthermore, the human-computer interaction module supports temporary adjustments to the system's operating mode via external commands. For example, when debugging equipment or handling special products, operators can connect external devices (such as computers or remote controls) to send commands to the system, switching it to debug mode or manual mode to perform special operations.

[0060] Using the data acquisition and analysis system provided in the embodiments of the present application, when processing a processed product, for example, cereal biscuits, the biscuits enter the detection area via a conveyor mechanism. The conveyor mechanism drives the biscuits forward at a speed of 0.3 m / s. When the biscuits reach the positioning sensor (located 15 cm in front of the image acquisition area), the sensor detects the object and sends a trigger signal to the data processing module, and the system enters the working state: The data processing module instructs the light source module to switch to visible light mode (5500K LED surface light source), and the light intensity is automatically adjusted to 3000 lux to ensure that the surface texture of the biscuits is clearly visible; at the same time, the near-infrared light source is in standby mode, waiting for the spectrum collection phase to start; A horizontal camera captures the top of the biscuit to detect cracks with a diameter of 0.5mm or more. A 45° tilted camera captures the side to detect edge deformation of 1mm or more. After visible light acquisition is complete, a near-infrared light source (900-1700nm band) is activated to collect spectral data of the biscuit, focusing on analyzing moisture content (corresponding to the characteristic peak at 1940nm) and oil distribution (corresponding to the characteristic peak at 1720nm). The ingredient-aware module synchronously collects the environmental temperature and humidity (25°C, 45%RH) and the weight of the biscuit (8±0.2g). The specification adaptation module extracts the circular profile (diameter 5cm, thickness 0.8cm) of the biscuit through image extraction, matches it with the built-in "cereal biscuit" parameter model, and calls the corresponding detection parameters: the image acquisition accuracy is set to 100dpi, the spectral analysis range focuses on 900-1940nm, and the area is greater than or equal to 20cm 2 and the shape factor is greater than or equal to 0.9.

[0061] The data processing module receives the following data and performs calculations: Image features: no cracks on the top, flat edges on the sides (shape distortion ≤0.5mm); Spectral features: moisture content 5.2% (standard range 4%-6%), uniform oil distribution (deviation ≤5%); Environmental parameters: temperature and humidity are within the normal range (20-30°C, 40%-50%RH), and the weight meets the standard; Through the feature fusion algorithm, the comprehensive score (89 points, threshold 60 points) is calculated, and the product is determined to be good, and the count is accumulated +1.

[0062] The touch display screen displays the detection result of the biscuit in real time: "good product (moisture 5.2%, weight 8.1g)", and the cumulative good product number is updated from "n / 150" to "n+1 / 150"; the voice interaction unit synchronously broadcasts "detection qualified, continue to convey", and the operator can check the real-time good product rate of the batch through the screen.

[0063] Through the above data acquisition and analysis system, the integration of comprehensive detection, automatic adaptation and efficient process control is realized, the quality is accurately controlled through multi-dimensional data acquisition, the appearance details are captured through high-definition image capture, the internal composition is analyzed through spectral analysis, the temperature and humidity and dust are monitored through environmental sensing, compared with the traditional single detection method, the detection coverage and accuracy are effectively improved; secondly, the specification adaptation module has built-in multi-category parameter model, which can automatically identify different products such as biscuits and bread and call corresponding detection standards, cooperate with dynamic threshold generation technology, adapt to batch differences in real time, effectively reduce the parameter debugging time during changeover, meet the multi-variety small-batch production demand; and, the present application greatly reduces manual intervention through the whole process, effectively reduces the labor cost of production line, improves the counting accuracy, avoids the error of quantitative packaging, and effectively improves the intelligent level and production efficiency of grain processing.

[0064] Reference Figure 2 In some embodiments, the data processing module includes a deep learning fusion module, a dynamic threshold generation module, and a counting compensation algorithm module.

[0065] The deep learning fusion module is used to fuse the appearance and morphology image with the component distribution spectrum data, combine the processing environment parameters, and construct a comprehensive evaluation dimension containing physical characteristics and component characteristics; the deep learning fusion module is used to deeply fuse the appearance and morphology image acquired by the image acquisition module with the component distribution spectrum data acquired by the spectral imaging unit, and combine the processing environment parameters acquired by the component perception module to construct a more comprehensive and accurate comprehensive evaluation dimension. The model is trained based on a deep learning algorithm through a large amount of sample data. The training samples include image data, spectrum data, corresponding environment parameters and quality determination results of grain processing products of various types and various quality states. Through training, the model can learn the internal relationship between different characteristics and product quality, and has strong feature extraction and pattern recognition capability.

[0066] In the model, the appearance and morphology image and the component distribution spectrum data are first extracted respectively. The shape distortion, surface cracks, color uniformity and other physical characteristics of the product are extracted from the appearance image; the moisture content, oil distribution and additive concentration of the product are analyzed from the spectrum data. Then, the model uses a specific fusion strategy to fuse these characteristics, and introduces environmental parameters as auxiliary information to comprehensively consider the influence of various factors on product quality.

[0067] For example, when detecting bread, the model will fuse and analyze the surface color of the bread (appearance characteristic), the internal moisture distribution (component characteristic) and the environmental temperature during baking (environmental parameter). If the surface color of the bread is too dark, the internal moisture content is too low, and the environmental temperature during baking is too high, the model will judge that the bread has been over-baked and determine it as a defective product.

[0068] The deep learning fusion module automatically learns and optimizes the feature fusion, and as the number of samples increases and the use time prolongs, the recognition accuracy and stability of the model are continuously improved, which can better adapt to product detection under different batches and different production conditions, and improve the analysis and detection accuracy and applicability.

[0069] The dynamic threshold generation module is used to automatically adjust the good product determination threshold according to the batch characteristics of the processed products after acquiring the detection parameters of the specification adaptation module. The dynamic threshold generation module can automatically adjust the standard of good product determination according to the batch characteristics of the processed products and the real-time production situation, and improve the accuracy and adaptability of the determination. In the traditional detection system, the good product determination threshold is usually fixed, and misjudgment may occur for products under different batches and different production conditions. The dynamic threshold generation module adjusts the threshold in combination with the batch characteristics of the processed products after acquiring the detection parameters of the specification adaptation module.

[0070] Batch characteristics include the origin of raw materials, the proportion, the running state of processing equipment, the change of production environment and other factors. For example, the moisture content standards of the final products of biscuits of the same brand should be different due to the differences in the moisture content of raw materials of different batches. The dynamic threshold generation module calculates the characteristic distribution of the products of the current batch by analyzing the initial sample data of the products of the current batch, such as the detection results of the previous products, and then adjusts the good product judgment threshold.

[0071] Meanwhile, the dynamic threshold generation module corrects the threshold according to the real-time changes in the production process. For example, after the equipment is used for a period of time, slight wear occurs, which causes slight changes in some characteristics of the products. The dynamic threshold generation module can capture these changes and make corresponding adjustments to the threshold to avoid a large number of product misjudgments due to slight changes in the equipment.

[0072] The generation process of the dynamic threshold adopts a statistical analysis method to ensure the rationality and reliability of the threshold through the analysis of historical data and real-time data. For example, for a batch of biscuits, the average value and the standard deviation of the moisture content of the first 50 products are calculated, and then the qualified range (threshold) of the moisture content is set according to the statistical law to ensure that most of the qualified products can pass the detection, and at the same time, the unqualified products can be accurately screened out.

[0073] The counting compensation algorithm module is used to eliminate the counting errors of the processing products in the overlapping and inclined states. Specifically, the counting compensation algorithm module is used to solve the counting error problem of the processing products in the overlapping and inclined states in the conveying process to ensure the accuracy of the counting results.

[0074] In actual production, the processing products may overlap or tilt on the conveying belt, which may cause misjudgment when the image acquisition module counts, such as mistaking two overlapping biscuits as one or counting one tilted biscuit as multiple. The counting compensation algorithm module analyzes the collected images, identifies these special states, and makes corresponding compensation corrections.

[0075] For the products in the overlapping state, the algorithm analyzes the contours and gray scale changes of the products in the images to determine the overlapping area and the overlapping degree between the products. Then, according to the normal size and shape of the products, the actual number of products in the overlapping area is estimated. For example, when two partially overlapping biscuits are detected, the algorithm identifies the edge features of the two biscuits to determine that they are two overlapping biscuits rather than one deformed biscuit, thereby increasing the number by 1 when counting.

[0076] For products in an inclined state, the algorithm determines the degree of inclination of the product by detecting the angle between the main axis of the product and the conveying belt. Then, according to the actual size of the product and the inclination angle, the relationship between the projection area of the product in the image and the actual area is calculated, and the counting result is corrected. For example, the projection area of an inclined biscuit in the image is smaller than the actual area, and the algorithm will correct the projection area according to the inclination angle to ensure that the counting will not be missed due to the reduction in area.

[0077] Through the counting compensation algorithm module, the system can effectively eliminate the counting errors caused by poor product placement, improve the accuracy of counting, and provide reliable quantity basis for subsequent packaging, sub-packaging and other links.

[0078] In some embodiments, the deep learning fusion module adopts a dual-branch network structure with an attention mechanism, which includes a first branch and a second branch, respectively responsible for extracting features of appearance morphology images and ingredient distribution spectrum data. The two branches work in parallel and each focuses on different types of feature extraction.

[0079] The first branch is used to extract shape distortion, surface crack, and color uniformity features in the appearance morphology image, i.e., the first branch mainly processes appearance morphology images, and its network structure is optimized and designed to effectively extract physical features such as shape distortion, surface crack, and color uniformity in the image. This branch adopts a convolutional neural network (CNN) structure, which gradually extracts low-level features (such as edges, textures) and high-level features (such as overall shape, defect area) of the image through multiple convolution and pooling operations. For example, when extracting surface crack features, the shallow convolution layer of the network is responsible for detecting edge information in the image, and the deep convolution layer combines these edge information to identify features such as crack location, length, and width.

[0080] The second branch is used to analyze the moisture content, oil distribution, and additive concentration characteristics in the ingredient distribution spectrum data, i.e., the second branch focuses on analyzing ingredient distribution spectrum data to extract ingredient features such as moisture content, oil distribution, and additive concentration. The second branch adopts a network structure suitable for processing sequence data, such as a recurrent neural network (RNN) or a fully connected network, which can analyze the response of spectrum data at different wavelengths. Spectrum data reflects the absorption of different wavelengths of light by internal ingredients of the product, and different ingredients have unique absorption peaks at specific wavelengths. For example, water has a specific absorption peak in the near-infrared band, and the second branch calculates the moisture content in the product by analyzing the intensity and position of this absorption peak in the spectrum data.

[0081] The attention mechanism automatically allocates feature weights of the two branches according to the processing product type. Different types of grain processing products have different appearance features and ingredient features that affect the quality. For example, for biscuit products, the integrity of the appearance (such as whether there are cracks, whether the shape is regular) has a greater impact on product quality; and for bread products, the internal ingredient distribution (such as moisture content, pore distribution) is more important. The attention mechanism can dynamically adjust the weights of the appearance features and ingredient features in the comprehensive evaluation according to the product type.

[0082] When detecting biscuits, the attention mechanism increases the weight of the first branch (appearance features), so that the model pays more attention to the surface cracks, shape distortion and other appearance defects of the biscuits; when detecting bread, the weight of the second branch (ingredient features) is increased, and the ingredient information such as moisture content and protein distribution of the bread is analyzed.

[0083] The advantage of the double-branch network structure is that it can adopt the most suitable feature extraction method for different types of data, improving the efficiency and accuracy of feature extraction. The two branches process image and spectral data respectively, avoiding interference between different types of data. The implementation of the attention mechanism is based on the identification of product types and the evaluation of feature importance. The model learns the correlation between appearance features and ingredient features and quality determination results in different product types, and automatically determines the importance weights of each feature. In feature fusion, the features of the two branches are weighted and combined according to these weights to form the final comprehensive features for quality determination. Through the attention mechanism, the deep learning fusion module can more intelligently adapt to the detection needs of different types of products, improving the accuracy and reliability of quality determination.

[0084] Reference Figure 3 In some embodiments, the dynamic threshold generation module includes a data feature extraction unit, a threshold calculation unit, and a threshold updating unit.

[0085] The data feature extraction unit is used to extract key feature data such as raw material ratio, processing temperature curve, and molding pressure from the real-time collected processing product batch, and to provide a basis for threshold calculation by extracting key feature data from the real-time collected processing product batch.

[0086] The data feature extraction unit extracts the proportion information of each raw material by obtaining the raw material feeding records in the production process, then extracts the curve features of temperature change over time by collecting temperature sensor data of the processing equipment, and extracts the average value, maximum value and fluctuation range of the molding pressure by collecting pressure sensor data of the molding equipment. In addition, the module also extracts the initial detection features of the products, such as the appearance and composition detection results of the previous products, as a supplement to the batch features. The extracted key feature data is standardized to eliminate the dimensional differences between different features, facilitating subsequent threshold calculation.

[0087] The threshold calculation unit is used to establish a threshold calculation model based on machine learning algorithm by analyzing historical qualified batch data, and calculate the good product judgment threshold of the current batch combined with real-time feature data. The historical qualified batch data contains a large amount of feature data and corresponding threshold information of different batches of qualified products, which is the basis for establishing the threshold calculation model. By learning the relationship between features and thresholds in historical data, the appropriate threshold can be predicted according to the new batch features.

[0088] When calculating the threshold of the current batch, the module inputs the real-time feature data provided by the data feature extraction unit into the threshold calculation model, and the model outputs the corresponding threshold according to the learned rules. For example, for a batch of biscuits, the proportion of oil in the raw materials is slightly higher than the historical average level. Since the higher oil content allows the product to have slightly higher moisture content without affecting the shelf life, the threshold calculation model will appropriately increase the qualified threshold of moisture content according to the relationship between oil proportion and moisture content threshold in historical data.

[0089] The threshold calculation model uses a regression analysis algorithm to continuously map between features and thresholds, ensuring that the calculated threshold accurately reflects the characteristics of the current batch of products. At the same time, the model will be updated and optimized regularly according to new production data to improve the adaptability and accuracy of the model.

[0090] The threshold updating unit dynamically corrects the calculated threshold according to factors such as environmental parameter fluctuations and equipment operating state changes in the production process.

[0091] Sudden changes in environmental parameters such as temperature and humidity can cause temporary changes in product quality characteristics. For example, a sudden increase in workshop humidity will cause the surface moisture content of the product to increase slightly. At this time, if the original threshold is still used, it will misjudge the qualified products as unqualified products. The threshold updating unit monitors the changes in environmental parameters and adjusts the threshold appropriately when the change exceeds a certain range. For example, when the humidity increases, the threshold of moisture content is appropriately increased.

[0092] Changes in equipment operating status can also affect product quality. For example, equipment wear and aging of components can cause systemic changes in certain product characteristics. The module collects operating parameters such as motor speed, vibration frequency, and temperature to determine the equipment's operating status. If an abnormality in equipment operating status occurs, such as an increase in vibration frequency, it indicates a minor equipment malfunction, resulting in a slight deviation in the product's shape. In this case, the module appropriately relaxes the shape-related thresholds.

[0093] The correction process of the threshold update unit uses incremental adjustment to achieve a smaller adjustment range each time, avoiding excessive fluctuations in the threshold due to over-adjustment. At the same time, the module will record the adjustment history of the threshold to facilitate subsequent analysis and tracing.

[0094] Reference Figure 4 In some embodiments, the counting compensation algorithm module includes a three-dimensional reconstruction subunit, an overlap determination subunit, and a morphology correction subunit.

[0095] The 3D reconstruction subunit uses multi-angle images to generate the 3D outline of the processed product. The subunit uses two high-definition cameras at different angles to capture images of the same product, and then calculates the 3D coordinate information of the product based on the principle of stereo vision, and then reconstructs the 3D outline of the product.

[0096] The principle of stereo vision is similar to human binocular vision. It uses the positional relationship (baseline distance) between two cameras and the shooting angle to calculate the spatial distance between corresponding points in the image. For example, if two cameras capture a cookie from a horizontal angle and a 45-degree angle, the subunit compares corresponding points on the cookie's edge in the two images, calculates their 3D coordinates, and then connects these coordinates to form the cookie's 3D outline.

[0097] The 3D contour clearly reflects the product's actual shape, height, and spatial position, avoiding the distortion caused by perspective in 2D images. For example, a tilted cookie may appear shorter in a 2D image, but a 3D contour accurately reflects its actual length and tilt angle. The generated 3D contour data is then passed to the overlap determination subunit and the morphology correction subunit to determine whether the products overlap and perform tilt correction.

[0098] The overlap determination subunit determines the occlusion relationship between the processed products through the spatial coordinate calculation of the three-dimensional contour to determine whether there is overlap. The subunit analyzes the three-dimensional contour data of multiple products generated by the three-dimensional reconstruction subunit and calculates the spatial distance and relative position relationship between the products.

[0099] When the three-dimensional contours of two products intersect in space, i.e., partially overlap, the subunit determines that there is overlap. During the determination process, the volume and area of the overlapping region are calculated to evaluate the severity of the overlap. For example, if the three-dimensional contours of two biscuits partially overlap in space, and the volume of the overlapping region accounts for 20% of the volume of a single biscuit, it is determined to be a mild overlap; if the volume of the overlapping region accounts for 50% or more, it is determined to be a severe overlap.

[0100] For overlapping products, the subunit estimates the actual number of products in the overlapping region based on the normal size and shape of the products. For example, for two biscuits that are slightly overlapping, by analyzing the shape and size of the overlapping region, it is determined that two complete biscuits are overlapping, rather than a broken biscuit, so that two products are counted. The result of the overlap determination is transmitted to the counting module to correct the initial counting result, ensuring the accuracy of the count.

[0101] The morphological correction subunit corrects the projected area of the oblique processed product to achieve accurate counting. The subunit analyzes the three-dimensional contour of the product generated by the three-dimensional reconstruction subunit to determine the oblique angle of the product, and then corrects the projected area of the product in the image according to the oblique angle.

[0102] The oblique angle of the product refers to the angle between the main plane of the product and the plane of the conveyor belt, which affects the projected area of the product in the image. For example, a biscuit placed horizontally has a projected area equal to the actual area; when the biscuit is tilted at an angle, the projected area is smaller than the actual area, and if the count is based on the projected area, it will lead to a misjudgment.

[0103] The morphological correction subunit calculates the oblique angle of the product based on the three-dimensional contour data, and then uses the trigonometric relationship to calculate the proportional relationship between the projected area and the actual area to correct the projected area. For example, a biscuit has an actual area of 10 cm 2 and an oblique angle of 30 degrees, with a projected area of about 8.66 cm 2 The correction subunit will correct the projected area to 10 cm 2 The corrected area is used for counting determination to ensure that even if the product is tilted, the counting result accurately reflects the actual number of products. At the same time, the correction subunit also corrects the shape of the product to remove the shape distortion caused by tilting, facilitating subsequent appearance quality detection.

[0104] In some embodiments, the data acquisition and analysis system further comprises an adaptive control module connected to the data processing module to automatically adjust the exposure parameters of the image acquisition module and the spectral ratio of the light source module according to changes in environmental parameters; In some embodiments, the data processing module is further configured to statistically analyze the number of defective products in a preset time period, and send a warning signal through the human-machine interaction module when the defective product ratio exceeds a preset threshold.

[0105] The preset time period can be set according to the production situation, such as every hour, every shift or every batch of products. In this time period, the data processing module records the number of products determined as defective products, and at the same time records the types of defective products, such as appearance defects (cracks, deformation), component overproof (excessive moisture, excessive additives), etc. The calculation method of the defective product ratio is the ratio of the number of defective products to the total number of products detected in the time period, expressed in percentage. For example, in a certain hour, a total of 1000 products are detected, of which 30 are defective products, and the defective product ratio is 3%.

[0106] The statistical analysis result is updated in real time and displayed on the touch display screen of the human-machine interaction module, and the operator can check the current defective product situation at any time. At the same time, the system saves historical statistical data to form a quality trend chart, such as the change of defective product ratio in different time periods within a day, the distribution of defective product types of different batches of products, etc., to facilitate the analysis of the causes and development trend of quality problems.

[0107] When the defective product ratio exceeds the preset threshold, the system sends a warning signal through the human-machine interaction module. The warning signal can be in various forms such as visual and auditory, to ensure that the operator can detect it in time. The preset threshold can be set according to the quality requirements of the product and production experience, generally 3%-5%. When the defective product ratio exceeds the threshold, the touch display screen displays a prominent red warning icon and flashes to prompt, and displays the specific value of the defective product ratio and the main defective types on the screen; the auditory warning is realized through a buzzer or a voice interaction unit. The buzzer will emit a continuous or intermittent alarm sound, and the voice interaction unit will broadcast a prompt message such as "high defective product rate, please check the production process". For different degrees of overproof, the intensity and frequency of the warning signal are different, such as when the defective product ratio exceeds 1.5 times the threshold, the alarm sound will be more urgent, and the voice prompt will be more frequent.

[0108] In addition, the system can also send the warning signal to the mobile device of the manager, such as a mobile phone APP, through the network, so that the manager can timely understand the quality abnormal situation even if he is not in the production site. After the warning signal is sent, the warning will be lifted only when the defective product ratio falls below the threshold or the operator manually resets it Embodiment Two With reference to Figure 5 The second aspect of the present application provides a grain processing device, comprising: The conveying mechanism 100 is a belt conveyor, and the surface of the belt of the belt conveyor is provided with anti-skid patterns; the conveying mechanism 100 is responsible for conveying the processing products to be detected from the feeding end to various processing links, such as the detection area, the rejection area and the packaging area. The belt surface is provided with anti-skid patterns in the form of, for example, diamond patterns, which can increase the friction between the products and the belt, preventing the products from sliding, deviating or falling during the conveying process due to vibration, acceleration or deceleration.

[0109] The length and width of the conveyor are designed according to the production site and the size of the products, and generally the length is 5-10 meters and the width is 0.5-1 meter, which can meet the conveying needs of most grain processing products. The conveying speed can be adjusted through the control system, and the range is usually 0.1-1 meter / second, which can be set according to the detection speed, production efficiency and other requirements. For example, in the case of fast detection speed, the conveying speed can be increased to improve the production efficiency; when the detection accuracy requirement is high, the conveying speed can be reduced to ensure that the detection module has enough time to complete data acquisition and analysis.

[0110] The driving device of the conveyor adopts a variable frequency motor, which has the advantages of smooth operation, wide speed regulation range and low energy consumption, and can ensure the stability of the conveying speed and reduce the influence of speed fluctuation on detection and counting.

[0111] As any one of the data acquisition and analysis systems for grain processing 200 provided in the embodiments, the image acquisition module, the light source module and the ingredient sensing module are arranged on the conveying path of the conveying mechanism 100.

[0112] The positioning sensor 300 is arranged on the conveying mechanism 100 and connected with the data acquisition and analysis system 200, and is used for positioning the position of the processing products and triggering the data acquisition and analysis system 200 to work. The positioning sensor 300 usually adopts a diffuse reflection type photoelectric sensor, which detects the presence of an object by emitting light and receiving reflected light. When the product moves with the conveying belt and passes through the positioning sensor 300, the product will reflect the light emitted by the sensor, and after the sensor receives the reflected light, it will output an electric signal to the data acquisition and analysis system 200, indicating that the product is about to enter the image acquisition area; the installation position of the sensor is about 10-20 centimeters in front of the image acquisition area, so that enough response time is left for the system to ensure that the image acquisition module, the light source module and the like can start working in time when the product reaches the center of the acquisition area. For example, when the conveying belt speed is 0.3 meters / second, the distance of 10 centimeters requires about 0.33 seconds of time, which is enough for the system to complete the start-up preparation.

[0113] The detection distance and sensitivity of the positioning sensor 300 can be adjusted to adapt to products of different sizes and different colors. For small products such as cereal, the sensitivity of the sensor can be increased; for products with darker colors, the intensity of the emitted light can be appropriately increased to ensure reliable detection.

[0114] The rejection mechanism 400 is provided on the conveying mechanism 100 and connected with the data acquisition and analysis system 200, and is used for rejecting defective products from the conveying mechanism 100. The rejection mechanism 400 mainly consists of a pneumatic push rod 410 and a defective product collection box 420. The pneumatic push rod 410 is installed on one side of the conveying belt, and the defective product collection box 420 is installed on the other side of the conveying belt, opposite the pneumatic push rod 410. When the data processing module determines that a product is defective, it will send a control signal to the rejection mechanism 400 and record the position of the product on the conveying belt. When the defective product moves to the position of the rejection mechanism 400 along the conveying belt, the pneumatic push rod 410 is driven by compressed air to quickly extend and push the defective product away from the conveying belt, so that it falls into the defective product collection box 420 on the opposite side.

[0115] The packaging mechanism 500 is connected with the data acquisition and analysis system 200. The packaging mechanism 500 includes a rack for placing packaging materials, a packaging forming assembly, and a sealing assembly to quantitatively package the good products according to the counting results.

[0116] The rack is used to place packaging materials such as packaging bags, packaging films, etc. It is usually designed with multiple storage locations to place different specifications of packaging materials at the same time, facilitating quick switching. Sensors are installed on the rack to detect the remaining amount of packaging materials. When the materials are insufficient, a signal will be sent to the data processing module to trigger a warning.

[0117] The packaging forming assembly is responsible for making the packaging materials into the required packaging shape, such as bag-shaped, box-shaped, etc. For packaging bags, the forming assembly folds and seals the flat packaging film to form the bag opening through mechanical devices; for packaging boxes, it forms the box shape by folding and bonding. The forming process is fast and efficient, and can match the speed of the conveying mechanism 100.

[0118] The sealing assembly is used to seal the packaging of the packaged products to ensure that the packaging is tightly sealed, preventing the products from getting wet, deteriorating, or falling apart. The sealing method is selected according to the type of packaging material, such as heat sealing, pressure sealing, and adhesion. For plastic packaging bags, heat sealing is usually used, which melts and bonds the sealing part of the packaging bag together by heating; for paper packaging, pressure sealing or adhesion is used.

[0119] The working flow of the packaging mechanism 500 is matched with the counting result of the data processing module. When the count reaches the preset packaging quantity (such as 10, 20), the data processing module sends a signal to the packaging mechanism 500, and the packaging mechanism 500 starts working: taking the packaging material from the rack, making the packaging through the forming assembly, then sending the corresponding number of good products into the packaging, and finally completing the sealing by the sealing assembly to form the final packaging product. The packaged products will be transported to the next link, such as boxing, stacking, etc.

[0120] In some embodiments, the rejection mechanism 400 includes a pneumatic push rod 410 and a defective product collection box 420, which are respectively arranged on both sides of the conveying mechanism 100, for pushing the defective product processing products into the defective product collection box 420.

[0121] The pneumatic push rod 410 is the execution component of the rejection mechanism 400, which adopts a double-acting cylinder, that is, it can be driven to extend and retract by compressed air, ensuring the flexibility and reliability of the action. The cylinder diameter and piston rod diameter are determined according to the required thrust and stroke. Generally, the cylinder diameter is 25-50 mm, and the piston rod diameter is 10-20 mm, which can provide sufficient thrust (such as 50-100 N) and stroke (such as 50-100 cm) to meet the rejection needs of products of different sizes.

[0122] In order to adjust the action speed of the pneumatic push rod 410, a one-way throttle valve is installed on the cylinder, and by adjusting the opening of the throttle valve, the extension and retraction speed of the push rod can be controlled. For example, for fragile products, the extension speed can be slowed down to avoid excessive impact; for fast-moving products, a faster extension speed is required to ensure timely rejection; a magnetic switch is also installed on the cylinder to detect the position of the piston rod, that is, whether the push rod is fully extended or retracted. The magnetic switch feeds back the position signal to the control system, and the control system judges whether the rejection action is completed according to the signal to ensure the accuracy of the action. If it is detected that the pneumatic push rod 410 is not fully extended or retracted, the system will issue a fault prompt for timely maintenance.

[0123] The defective product collecting box 420 is made of stainless steel, and a guide plate is arranged at the entrance of the defective product collecting box 420. The guide plate is in an inclined state (generally at an angle of 30-45 degrees with the horizontal plane), which can guide the defective product pushed out by the pneumatic push rod 410 to smoothly enter the collecting box, so as to avoid the product from falling outside the collecting box. The surface of the guide plate is smooth, so as to reduce the friction between the product and the guide plate, and ensure that the product can smoothly slide. A slope is arranged at the bottom of the box, so as to make the product gather in the box, and fully utilize the collecting space. A buffer pad such as a foam pad or a rubber pad can be arranged in the box. The thickness of the buffer pad is selected according to the weight and hardness of the product, so as to effectively absorb the impact force of the falling product, and prevent the product from being broken and generating noise. The defective product collecting box 420 can adopt a drawer type or a flip cover type structure, so that the operator can conveniently take out and process the defective product.

[0124] A diffuse reflection photoelectric sensor is further arranged on the defective product collecting box 420. The detection head of the sensor is aligned with the box. When the defective product is accumulated to a certain height (for example, reaches 80% of the volume of the collecting box), the sensor is blocked, and a signal is sent to the data processing module. The data processing module sends a prompt information "the defective product collecting box 420 is full, please clean it in time" through the man-machine interaction module.

[0125] Example three With reference to Figure 6 , the third aspect of the present application is a grain processing method applied to the grain processing device of example two, for counting, packaging and good product identification of the processed product, which comprises the following steps: S1: conveying the processed product to be processed at a set speed through the conveying mechanism; The processed product to be processed is conveyed at a set speed through the conveying mechanism. Before starting the conveying mechanism, the operator needs to set a suitable conveying speed according to the type of the processed product and the production plan. The setting of the conveying speed needs to consider the processing capacity of the subsequent detection, rejection and packaging links, so as to ensure the coordination between the links, for example, for biscuits and other small and thin products, the conveying speed can be set to 0.3 m / s, which can ensure that the detection module has enough time to complete image acquisition and analysis, and can also meet a certain production efficiency; for bread and other larger products, the conveying speed can be appropriately reduced to 0.2 m / s, so as to more clearly acquire images and perform component analysis.

[0126] After the conveying mechanism is started, the processed products to be processed are evenly placed at the feeding end of the conveying belt, and the conveying belt drives the products to move forward. During the conveying process, the products can maintain a stable position and will not slide or deviate significantly due to the anti-skid pattern on the surface of the belt. The operator needs to ensure that the products are evenly distributed on the conveying belt to avoid excessive accumulation, which may affect the accuracy of subsequent detection and counting. During the operation of the conveying mechanism, its speed is monitored in real time. If abnormal fluctuations occur, such as sudden deceleration or acceleration, the system will issue a warning to remind the operator to check whether there is a fault in the equipment, such as a motor problem or belt slip.

[0127] S2: When the positioning sensor detects that the processed product reaches the image acquisition area, the image acquisition module, the light source module, and the component sensing module are triggered to work, and multiple information data of the processed product are collected and transmitted to the data processing module; When the positioning sensor detects that the processed product reaches the image acquisition area, the image acquisition module, the light source module, and the component sensing module are triggered to work, and multiple information data are collected and transmitted to the data processing module. The positioning sensor is installed in front of the image acquisition area. When the product moves to the detection range of the sensor with the conveying belt, the sensor detects the presence of the product and immediately sends a trigger signal to the data processing module. After receiving the signal, the data processing module issues a start instruction to the image acquisition module, the light source module, and the component sensing module. After receiving the instruction, the image acquisition module simultaneously captures the product from two different angles using two high-definition cameras to obtain the appearance image of the product. The light source module automatically switches to the appropriate light source (visible light or near-infrared light) according to the collection requirements and adjusts the light intensity to provide good lighting conditions for image acquisition and spectral analysis. The spectral imaging unit is simultaneously started to collect the component distribution spectral data of the product. After receiving the instruction, the component sensing module simultaneously collects parameters such as temperature, humidity, and dust concentration of the processing environment, as well as the weight information of the product. These parameters and data are collected within a very short time (usually within 0.1 seconds) and are sent in real time to the data processing module through the data transmission line.

[0128] To ensure the accuracy of the collected data, before triggering the collection, the system will perform self-checking on each module, such as whether the camera is working normally, whether the light source is on, and whether the sensor is sensitive, etc. If an abnormality is found, a fault prompt will be issued and the collection process will be paused.

[0129] S3: The data processing module analyzes the multiple information data to determine whether the processed product is a good product, and simultaneously uses a counting algorithm to count the good products and sends the counting results and good product identification results to the packaging mechanism and the rejection mechanism, respectively; The data processing module analyzes the collected information data to determine whether the processed product is a good product, counts the good products, and sends the results to the corresponding mechanism. The data processing module first preprocesses the received image data, such as removing noise and enhancing contrast, to make the product features in the image more clear. Then, the data processing module extracts and fuses the features of the preprocessed image and spectral data through a deep learning fusion module, and constructs a comprehensive evaluation dimension combined with environmental parameters. Based on the comprehensive evaluation, the data processing module compares the features of the product with the good product judgment threshold calculated by the dynamic threshold generation module to determine whether the product is a good product. For example, if the surface crack length of the product is less than the threshold value and the moisture content is within the threshold range, it is determined to be a good product; otherwise, it is determined to be a defective product.

[0130] While determining the good product, the counting compensation algorithm module counts the good products. The algorithm identifies and compensates for overlapping and tilted products by analyzing the number, position, and state of the products in the image, and obtains accurate counting results. For example, when two overlapping good biscuits are detected, the counting compensation algorithm module will identify that there are actually two products, and the counting result will increase by 2. The data processing module sends the counting result to the packaging mechanism in real time, so that the packaging mechanism can accumulate the count, and when the preset packaging quantity is reached, the packaging operation is started. At the same time, the good product recognition result is sent to the rejection mechanism, and if it is a defective product, the position of the product is marked so that the rejection mechanism can remove it at the appropriate time.

[0131] S4: If it is identified as a defective product, the rejection mechanism removes it when the processed product reaches the rejection position; if it is identified as a good product, the conveying mechanism conveys it to the packaging mechanism; For products determined to be defective, the data processing module continuously tracks the position of the defective product on the conveyor belt (based on the speed and movement time of the conveyor belt). When the defective product reaches the position of the rejection mechanism, the data processing module sends a rejection signal to the rejection mechanism. The pneumatic push rod of the rejection mechanism immediately extends to push the defective product away from the conveyor belt and into the defective product collection box. The push rod retracts quickly after completing the rejection action, waiting for the arrival of the next defective product.

[0132] For products determined to be good, the conveying mechanism continues to convey them forward, and they will not be disturbed when passing through the rejection mechanism, smoothly entering the subsequent packaging area. During the conveying process, the system monitors the position of the good product to ensure that it can accurately reach the feeding port of the packaging mechanism; to ensure the accuracy of the screening, the system verifies the rejection action, such as a secondary detection (simple image recognition) after the rejection mechanism to confirm that the defective product has been successfully removed. If a defective product is found that has not been removed, an alarm will be issued and recorded to facilitate the operator to investigate the cause. At the same time, for good products, a simple position detection is also performed to ensure that they do not deviate from the conveying path.

[0133] S5: The packaging mechanism uses the packaging forming assembly and the sealing assembly to package the corresponding number of good quality processed products according to the received counting result when the counting reaches the set number.

[0134] The control system of the packaging mechanism receives the counting result sent by the data processing module in real time and accumulates it. When the accumulated counting reaches the preset packaging number (such as 10 biscuits per package), the packaging mechanism starts the packaging process. First, the packaging forming assembly takes out the corresponding packaging material (such as a packaging bag) from the rack, and expands or folds the packaging material into the required packaging shape through mechanical devices, then the conveying mechanism sends the corresponding number of good quality products into the formed packaging. For small and regular products such as biscuits, the products are usually pushed into the packaging through a conveyor belt; for larger or irregular products, a special pushing mechanism is needed to assist.

[0135] After the products are loaded into the packaging, the packaging is conveyed to the sealing assembly, which seals the packaging according to the type of packaging material. For example, plastic packaging bags are sealed by heating and pressing the bag opening with a heating device; paper packaging is sealed by pressing or gluing. The sealing process is fast and efficient, ensuring firm and well-sealed sealing. After sealing, the packaged products are conveyed out of the packaging mechanism and into the next step, such as manual boxing or automatic stacking. At the same time, the counting of the packaging mechanism is cleared and a new counting is started, waiting for the next time the number reaches the packaging number to repeat the above packaging process.

[0136] In some embodiments, the cereal processing method further comprises S6: When it is detected that the packaging material is insufficient, a signal is sent to the data processing module to control the conveying mechanism to pause conveying, and a warning signal is issued through the human-computer interaction module.

[0137] The packing mechanism is provided with a packaging material detection sensor, such as an infrared sensor or an ultrasonic sensor, which continuously monitors the remaining amount of packaging material on the rack. When the remaining amount of packaging material is lower than a preset threshold (e.g., only enough for 10 packs of packaging needs), the sensor sends a signal to the data processing module. Upon receiving the signal of insufficient packaging material, the data processing module immediately sends a warning message through the human-computer interaction module, such as displaying "insufficient packaging material, please replenish in time" on the touch display screen, and the voice interaction unit also gives a corresponding voice prompt. To prevent the product from continuing to be transported and causing accumulation during the replenishment of materials, the data processing module sends a pause signal to the conveying mechanism, which stops running. After receiving the warning, the operator replenishes the packaging material on the rack in time. After the replenishment is completed, the operator sends a signal to the system through the "continue" button of the human-computer interaction module, and the data processing module controls the conveying mechanism to resume running after receiving the signal. The packaging mechanism also continues the packaging operation, and the warning signal is removed. The packaging material warning function ensures the continuity of production, reduces downtime and product waste caused by insufficient materials, and improves production efficiency.

[0138] The data acquisition and analysis system for grain processing, the processing device and the processing method provided by the application realize the deep integration of omnibearing detection, automatic adaptation and efficient process control.

[0139] Firstly, multi-dimensional data acquisition realizes accurate quality control. High-definition image capture, spectral analysis of internal components, and environmental sensing monitoring of temperature, humidity and dust are used to build a "appearance-component-environment" three-dimensional evaluation system. Compared with traditional single detection methods, this system can identify both overt and latent defects such as cracks and excessive moisture, and improve the detection coverage.

[0140] Secondly, the intelligent adaptation capability greatly improves the production flexibility. The specification adaptation module is built-in with multi-category parameter models, which can automatically identify different products such as biscuits and bread and call corresponding detection standards. Combined with dynamic threshold generation technology, it can adapt to batch differences in real time, greatly shortening the parameter debugging time during production changeover and meeting the needs of multi-variety small-batch production.

[0141] Thirdly, full-process automation reduces manual intervention. From the triggering of the positioning sensor to data acquisition, to the determination of good products by the data processing module, to the elimination of errors by the counting compensation algorithm module, to the linkage operation of the rejection mechanism and the packaging mechanism, a closed loop of "detection-selection-packaging" is formed, which reduces the labor cost of a single production line and effectively improves the counting accuracy, avoiding quantitative packaging errors.

[0142] In summary, the application solves the problems of low counting accuracy, incomplete good product identification and poor adaptability in the existing grain processing process. By integrating multi-dimensional data acquisition, intelligent analysis and automatic control functions, the intelligent level and production efficiency of grain processing are effectively improved.

[0143] The above description is merely a preferred embodiment of the present application and is not intended to limit the present application. Various modifications and variations are possible for those skilled in the art. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present application shall be included within the scope of protection of the present application.

[0144] It will be apparent to those skilled in the art that the present application is not limited to the details of the exemplary embodiments described above, and that the present application can be implemented in other specific forms without departing from the spirit or essential characteristics of the present application. Therefore, the embodiments should be considered in all respects as illustrative and non-restrictive, and the scope of the present application is defined by the appended claims, not the foregoing description, and all variations within the meaning and range of equivalents of the claims are intended to be included therein. Any reference sign in a claim should not be construed as limiting the claim to which it relates.

Claims

1. A data collection and analysis system for grain processing, characterized in that: include An image acquisition module comprising at least two high-definition cameras set at different angles and a spectral imaging unit, wherein the high-definition cameras are used to acquire images of the appearance of the processed product, and the spectral imaging unit is used to acquire spectral data of the component distribution of the processed product; A light source module, configured in conjunction with the image acquisition module, includes a switchable visible light source and a near-infrared light source, and is used to provide adaptive lighting for different types of data acquisition; The component sensing module includes a temperature and humidity sensor and a weight sensor to collect processing environment parameters; The specification adaptation module has built-in parameter models for various processed products of different specifications. It can automatically match the corresponding detection parameters by identifying the initial features of the processed products, including image acquisition accuracy, spectral analysis range, and counting judgment criteria. A data processing module is connected to the image acquisition module, the light source module, and the component perception module, and is used to perform feature fusion of the appearance morphology image and the component distribution spectrum data, and to construct a comprehensive evaluation dimension including physical characteristics and component characteristics in combination with the processing environment parameters, so as to determine whether the processed product is good and count the good products; The human-computer interaction module includes a touch screen and a voice interaction unit, which is used for operators to manually input parameters, query detection data and receive voice prompt information from the system, and supports temporary adjustment of the system's operating mode through external commands.

2. The data acquisition and analysis system according to claim 1, characterized in that: The data processing module includes: A deep learning fusion module is used to fuse the appearance morphology image with the component distribution spectrum data, and combine it with the processing environment parameters to construct a comprehensive evaluation dimension that includes physical and component characteristics; The dynamic threshold generation module automatically adjusts the good product judgment threshold based on the real-time collected batch characteristics of the processed products after obtaining the detection parameters of the specification adaptation module; The counting compensation algorithm module is used to eliminate counting errors when the processed products are overlapped or tilted.

3. The data acquisition and analysis system according to claim 2, characterized in that: The deep learning fusion module adopts a dual-branch network structure with an attention mechanism, including: The first branch is used to extract shape distortion, surface cracks, and color uniformity features from the appearance morphology image; The second branch is used to analyze the moisture content, oil distribution and additive concentration characteristics in the component distribution spectrum data; The attention mechanism automatically assigns feature weights to the two branches according to the type of processed products.

4. The data acquisition and analysis system according to claim 2, characterized in that: The dynamic threshold generation module includes: Data feature extraction unit, used to extract key feature data such as raw material ratio, processing temperature curve, molding pressure, etc. from the processed product batches collected in real time; The threshold calculation unit, based on a machine learning algorithm, establishes a threshold calculation model by analyzing historical qualified batch data and calculates the qualified product threshold for the current batch in combination with real-time feature data; The threshold update unit dynamically corrects the calculated threshold according to factors such as environmental parameter fluctuations in the production process and changes in equipment operating status.

5. The system according to claim 2, wherein: The counting compensation algorithm module includes: The 3D reconstruction subunit generates the 3D contour of the processed product using multi-angle images; The overlap determination subunit determines the occlusion relationship between the processed products by calculating the spatial coordinates of the three-dimensional contours; The morphological correction subunit corrects the projected area of ​​the tilted processing products to achieve accurate counting.

6. The data acquisition and analysis system according to claim 1, characterized in that: The data processing module is also used to perform statistical analysis on the number of defective products within a preset time period, and to send out an early warning signal through the human-computer interaction module when the proportion of defective products exceeds a preset threshold.

7. A grain processing device, characterized in that: include: A conveying mechanism (100), wherein the conveying mechanism (100) is a belt conveyor, and the belt surface of the belt conveyor is provided with anti-slip patterns; The grain processing data acquisition and analysis system (200) according to any one of claims 1 to 6, wherein the image acquisition module, the light source module and the component perception module are arranged on a conveying path of the conveying mechanism; a positioning sensor (300), arranged on the conveying mechanism and connected to the data acquisition and analysis system (200), for locating the position of the processed product and triggering the operation of the data acquisition and analysis system (200); a rejecting mechanism (400), disposed on the conveying mechanism (100) and connected to the data acquisition and analysis system (200), for rejecting defective processed products from the conveying mechanism (100); A packaging mechanism (500) is connected to the data acquisition and analysis system (200), and the packaging mechanism (500) includes a material rack for placing packaging materials, a packaging forming component, and a sealing component, so as to quantitatively package the processed products of good quality according to the counting results.

8. The data acquisition and analysis system according to claim 7, characterized in that: The rejection mechanism (400) includes a pneumatic push rod (410) and a defective product collection box (420), and the pneumatic push rod (410) and the defective product collection box (420) are respectively arranged on both sides of the conveying mechanism (100) and are used to push defective processed products into the defective product collection box (420).

9. A grain processing method, applied to the grain processing device according to any one of claims 7-8, for counting, packaging and identifying good products of processed products, characterized in that: The following steps are involved: S1: The processed product to be processed is transported at a set speed through the conveying mechanism; S2: When the positioning sensor detects that the processed product has reached the image acquisition area, it triggers the image acquisition module, light source module, and component perception module to work, collect various information data of the processed product, and transmit them to the data processing module; S3: The data processing module analyzes various information data to determine whether the processed product is good or not. It also uses a counting algorithm to count the good products and sends the counting results and good product identification results to the packaging mechanism and the rejection mechanism respectively. S4: If the product is identified as defective, the rejection mechanism will reject it when it reaches the rejection position; if it is identified as good, the conveying mechanism will transport it to the packaging mechanism; S5: The packaging mechanism packages the corresponding number of good processed products using the packaging forming component and the sealing component according to the received counting result when the count reaches the set number.

10. The method for processing grains according to claim 9, wherein: Also includes S6: When it is detected that the packaging material is insufficient, a signal is sent to the data processing module to control the conveying mechanism to suspend conveying, and an early warning signal is issued through the human-computer interaction module.

Images