Commodity identification method and system based on picture classification model, and vending machine

Abstract

The invention relates to the technical field of commodity recognition, in particular to a commodity recognition method and system based on a picture classification model and a vending machine. The method comprises the following steps: obtaining commodity basic information data, and carrying out commodity category physical parameter analysis based on the data to obtain physical attribute data of a commodity; commodity category images are acquired according to the physical parameters, and contrast enhancement processing is performed on the images, so that the identifiability of the images is improved; carrying out commodity target detection by using the enhanced image data, extracting a commodity target area and calculating an image feature vector; analyzing fine features of the external form of the commodity, and performing model identification adjustment on feature data in combination with a picture classification model; and based on the adjusted classification model, carrying out classification identification processing on the basic information of the commodity to obtain an accurate commodity classification result. According to the method, commodity image recognition is optimized, so that commodity image recognition and classification are more accurate and efficient.

Description

Technical Field

[0001] The present invention relates to the technical field of commodity identification, and in particular to a commodity identification method, system and vending machine based on an image classification model. Background Art

[0002] Commodity recognition methods based on image recognition technology, especially those based on image classification models, have gradually become a more intelligent and efficient solution. Image classification models can extract rich visual features from commodity images through deep learning technology, thereby realizing automatic recognition and classification of commodities. In particular, deep learning models such as convolutional neural networks (CNNs) can be trained on the basis of large-scale commodity image data and have powerful image feature learning and extraction capabilities. Commodity recognition methods based on image classification models can automatically identify commodity categories, brands, specifications and other information by collecting and analyzing commodity images, and can quickly and accurately identify commodities by comparing images with commodity data in the database. In practical applications, especially in scenarios such as unmanned vending machines, smart retail systems and automated warehouse management, image recognition technology can greatly improve operational efficiency, reduce manual intervention, and reduce error rates. However, a traditional commodity recognition method based on image classification models has the problem of inaccurate commodity recognition and inaccurate recognition of abnormal commodity category images. Summary of the invention

[0003] Based on this, it is necessary to provide a commodity identification method, system and vending machine based on an image classification model to solve at least one of the above technical problems.

[0004] To achieve the above purpose, a commodity identification method based on an image classification model includes the following steps:

[0005] Step S1: Obtain basic information data of the commodity; perform commodity category physical parameter analysis based on the commodity basic information data to obtain commodity category physical parameter data;

[0006] Step S2: collecting commodity category images according to the commodity category physical parameter data, thereby obtaining commodity category image data; performing commodity image contrast enhancement processing on the commodity category image data, thereby obtaining commodity category image contrast enhanced data;

[0007] Step S3: performing commodity target detection according to the commodity category image contrast enhancement data, thereby obtaining commodity category image target data; performing commodity category image feature vector calculation according to the commodity category image target data, thereby obtaining commodity category image feature vector data;

[0008] Step S4: Analyze the subtle features of the external morphology of the commodity category based on the commodity category image feature vector data to obtain the subtle feature data of the external morphology of the commodity category; adjust the model recognition data of the subtle feature data of the external morphology of the commodity category based on the picture classification model to obtain the picture classification recognition adjustment model; perform commodity classification recognition processing on the commodity basic information data based on the picture classification adjustment model to obtain the commodity classification recognition data.

[0009] By implementing the above method, the present invention can effectively improve the accuracy and robustness of the commodity recognition system, obtain the basic information data of the commodity and perform physical parameter analysis, which can provide a reliable basis for subsequent image acquisition and processing, and ensure that the system can identify and distinguish different categories of commodities. The analysis results of the physical parameter data of the commodity category provide a valuable reference for the image acquisition process, help identify the size, weight and other information of the item, and these information can optimize the direction and method of image acquisition, so that the acquired image can more accurately reflect the appearance characteristics of the commodity. After the commodity image is acquired and processed by contrast enhancement, the image quality problems caused by factors such as ambient lighting and shooting angle are eliminated, and the clarity and details of the image are ensured, thereby effectively avoiding the influence of low-quality images on the subsequent recognition accuracy. Image contrast enhancement can not only improve the detail performance of the image, but also highlight the outline of the commodity, making the target detection more accurate, and providing high-quality data for subsequent image feature extraction. With the completion of commodity target detection, the commodity area in the image can be accurately identified and further processed. The commodity category image target data provides a clear basis for the subsequent image feature vector calculation, so that the unique features of the commodity can be extracted and quantified. The generation of feature vectors provides an objective basis for the identification of goods. These feature data can significantly improve the accuracy of commodity classification and avoid misjudgment caused by factors such as similar appearance or lighting. In combination with the picture classification model, the subtle features of the external morphology of the commodity category are deeply analyzed and the model is adjusted to further optimize the recognition accuracy. Through the adjusted classification model, the basic information data of the commodity is processed, the category of the commodity is accurately identified, and the subtle differences between different commodities are accurately distinguished, which not only improves the accuracy of commodity classification, but also enhances the adaptability of the system to commodity images in complex scenes, and improves the efficiency and accuracy of recognition. Through accurate commodity recognition, the entire system can achieve efficient and reliable automated commodity recognition and classification processing, providing more powerful technical support for applications. Therefore, the present invention is an optimization process made to a traditional commodity recognition method based on a picture classification model, which solves the problem of inaccurate commodity recognition in a traditional commodity recognition method based on a picture classification model, and the problem of inaccurate recognition of abnormal commodity category images, improves the accuracy of commodity recognition, and the accuracy of abnormal recognition of commodity category images.

[0010] Preferably, step S1 comprises the following steps:

[0011] Step S11: Obtain basic information data of the product;

[0012] Step S12: collecting commodity category and brand data according to commodity basic information data, thereby obtaining commodity category and brand data;

[0013] Step S13: performing commodity category material analysis according to commodity category brand data, thereby obtaining commodity category material data;

[0014] Step S14: measuring the commodity category size according to the commodity category brand data, thereby obtaining the commodity category size data;

[0015] Step S15: Perform commodity category physical parameter analysis based on commodity category size data and commodity category material data to obtain commodity category physical parameter data.

[0016] Through these steps, the present invention can comprehensively and accurately obtain and analyze the basic information of the commodity, provide a reliable basis for subsequent commodity identification, and obtain the basic information data of the commodity to help establish the preliminary archive of the commodity, laying the foundation for subsequent analysis. This information will be used to build a detailed physical parameter model of the commodity, provide key information such as brand, size and material, etc., to ensure that the commodity can be accurately classified. In the process of brand data collection, by analyzing the brand information of the commodity, the system is helped to quickly identify the category to which the commodity belongs, provide the necessary brand identification dimension, and further promote the accuracy of commodity classification. Brand information plays an important guiding role in subsequent material analysis and size measurement, which helps to improve the accuracy of material and size data analysis and avoid errors in these steps. Material analysis allows the physical properties of the commodity to be described in detail. The material characteristics have an impact on the appearance, weight and structure of the commodity. The clear analysis of these factors helps to deeply understand the basic composition and appearance of the commodity, and then affects the subsequent image acquisition and image processing links. Size measurement provides the system with accurate size data of the commodity, which is crucial for the identification of the commodity. Especially when it comes to the identification of large quantities of commodities, the size information effectively reduces the identification errors caused by the similarity of the commodity shape. At the same time, combined with material data, size information will further promote the comprehensive analysis of the physical parameters of the goods. The comprehensive analysis of physical parameters covers multiple aspects such as the weight, volume, density, etc. of the goods. This information provides clear technical guidance for the image acquisition, target detection and image feature analysis of the goods, helping the system to optimize the image acquisition method and processing path, thereby improving the accuracy and efficiency of product recognition.

[0017] Preferably, step S15 comprises the following steps:

[0018] Step S151: Calculate the volume of the commodity category according to the commodity category size data, thereby obtaining the commodity category volume data;

[0019] Step S152: collecting commodity category density according to commodity category material data, thereby obtaining commodity category density data;

[0020] Step S153: Calculate the commodity category quality according to the commodity category quality measurement data and the commodity category volume data, thereby obtaining the commodity category quality data;

[0021] Step S154: Calculate the commodity weight according to the commodity category mass data and the commodity category volume data to obtain the commodity category weight data;

[0022] Step S155: Detecting the surface roughness of the commodity according to the commodity category material data, thereby obtaining the commodity category surface roughness data;

[0023] Step S156: Perform commodity category physical parameter analysis based on commodity category surface roughness data and commodity category weight data to obtain commodity category physical parameter data.

[0024] The physical properties of the goods of the present invention can be accurately calculated and analyzed, thereby providing more detailed and accurate data support for subsequent identification and classification. The calculation of the volume of the commodity category is the first step in analyzing the volume of the commodity, which can provide an accurate measurement for the space occupation of the commodity and lay the foundation for the subsequent quality and weight analysis. The volume data not only helps to infer the appearance characteristics of the commodity, but also can form an effective comparison between different physical parameters, making the analysis of physical properties more comprehensive. Then, density acquisition provides crucial data support for subsequent quality and weight measurement. The density of the commodity is closely related to its material and overall structure, which can effectively improve the accuracy of commodity identification, especially when the commodity material is more complex or the form is diverse. Through mass calculation, combined with volume data, the quality of the commodity can be accurately estimated, which is crucial for the actual transportation, storage and subsequent management of the commodity, and also provides a basis for optimizing the recognition accuracy in the image processing step, especially in some model training related to physical properties. The measurement of weight further promotes the precision of commodity attributes, which is one of the actual performances of commodities in different environments, can reflect the actual use experience of the commodity, and provide data support for consumer experience and operating system. The detection of product surface roughness provides a key parameter for detailed analysis of product appearance. This test can reveal the impact of product surface details on light reflection, which in turn affects the image acquisition quality of the product, ensuring that details will not be misidentified due to surface irregularities during image analysis.

[0025] Preferably, step S2 comprises the following steps:

[0026] Step S21: collecting commodity category images according to commodity category physical parameter data, thereby obtaining commodity category image data;

[0027] Step S22: performing commodity image preprocessing on the commodity category image data, thereby obtaining commodity category image preprocessing data;

[0028] Step S23: performing commodity category image anomaly detection based on commodity category image preprocessing data and commodity category physical parameter data to obtain commodity category image anomaly data;

[0029] Step S24: performing product image contrast enhancement processing on the product category image abnormal data to obtain product category image contrast enhanced data.

[0030] Through these steps, the image data of the commodity is systematically processed from acquisition to enhancement, ensuring that the performance of the commodity in the visual recognition process reaches the best effect. The commodity category image acquisition is carried out through the commodity category physical parameter data, which can ensure that the image is accurately recorded according to the actual physical characteristics of the commodity during the acquisition stage, thereby providing more realistic original data for subsequent image analysis. Image preprocessing effectively removes noise, corrects distortion, and performs common operations such as background removal and brightness adjustment on the image, thereby ensuring the quality of the commodity image data and avoiding the influence of external interference on the recognition effect. This step lays a more stable and clear foundation for subsequent analysis and feature extraction. Image anomaly detection based on commodity category image preprocessing data and commodity category physical parameter data can timely discover defects, distortion or irregularities in the image. The beneficial effect of this step is that it can effectively eliminate the problem of unstable image quality caused by external factors such as illumination, angle or reflection. Image anomaly detection provides a guarantee for improving the quality of commodity images, effectively avoiding the influence of inaccurate or distorted images on the accuracy of subsequent recognition, and commodity image contrast enhancement processing makes the commodity image more vivid and clear visually by strengthening the details in the image. Contrast enhancement can improve the visibility of details in product images, thereby improving product recognition accuracy, especially in conditions of low light or poor image quality, effectively improving the stability and effectiveness of image recognition.

[0031] Preferably, step S23 includes the following steps:

[0032] Step S231: Calculate the surface reflectivity of the commodity category according to the physical parameter data of the commodity category to obtain the surface optical refractive index data of the commodity category;

[0033] Step S232: performing commodity image quality instability analysis on the commodity category image preprocessing data according to the commodity category surface optical refractive index data to obtain commodity image quality instability analysis;

[0034] Step S233: performing commodity category scattering irregularity enhancement analysis according to the commodity category physical parameter data to obtain commodity category scattering irregularity enhancement data;

[0035] Step S234: performing detail loss estimation on the commodity category image preprocessing data according to the commodity category scattering irregularity enhancement data to obtain commodity category image detail loss data;

[0036] Step S235: performing commodity category image anomaly detection based on commodity category image detail loss data and commodity image quality instability analysis to obtain commodity category image anomaly data.

[0037] Through these steps, the quality of the commodity image is effectively controlled and optimized. The calculation of the reflectivity of the commodity category surface provides an important physical parameter basis for image quality analysis, and then the optical refractive index data of the commodity is inferred, which is crucial to understanding the way light interacts with the commodity surface. Next, the unstable quality analysis of the commodity image is performed based on the optical refractive index data of the commodity category surface, and the image quality fluctuation caused by the change of optical properties is effectively identified, thereby providing key data support for subsequent image adjustment. In addition, the scattering irregularity enhancement analysis can accurately reveal the risk of image detail loss caused by uneven light scattering on the commodity surface. This analysis helps to discover potential image quality problems in advance and avoid the impact of detail loss on subsequent recognition. The image preprocessing data is estimated for detail loss based on the scattering irregularity enhancement data, which provides accurate predictions for adjusting image quality and restoring lost details, ensuring the integrity of the commodity image. Image anomaly detection based on image detail loss data and image quality instability analysis can comprehensively identify and solve image anomalies caused by factors such as illumination and refraction, ensure that the collected commodity images are stable and clear, and provide more reliable data input for the commodity recognition model.

[0038] Preferably, step S232 includes the following steps:

[0039] According to the optical refractive index data of the commodity category surface, the commodity category refractive index is divided into the commodity category surface high refractive index data and the commodity category surface low refractive index data;

[0040] According to the high refractive index data of the commodity category surface, the change of the light propagation direction of the commodity surface is estimated to obtain the change data of the light propagation direction of the commodity surface;

[0041] The product category image distortion is estimated based on the data of the change in the light propagation direction on the product surface, and the product category image distortion data is obtained;

[0042] Based on the low refractive index data of the commodity category surface, the excessive reflection of the light of the commodity category is estimated to obtain the excessive reflection data of the light of the commodity category;

[0043] According to the product category light over-reflection data and the product category image distortion data, the product category image preprocessing data is estimated to obtain the product category image distortion data;

[0044] The commodity category image distortion data and the commodity category image deformation data are used to perform commodity image quality instability analysis to obtain commodity image quality instability analysis.

[0045] The present invention significantly improves the quality and stability of commodity images through fine division and analysis of the refractive index of the commodity surface. By dividing the optical refractive index data of the commodity surface into high and low refractive index, the optical characteristics of the commodity surface can be clearly distinguished, providing accurate basic data for subsequent light propagation analysis. For high refractive index areas, the change in light propagation direction is estimated, so that the potential impact of image distortion can be predicted, and the analysis of image distortion data provides key guidance for subsequent image correction. At the same time, the estimation of excessive light reflection in low refractive index areas helps to identify and suppress image quality loss caused by excessive light reflection, and reduce image distortion. Further combining the excessive light reflection data with the image distortion data for distortion estimation, the details of the image are better restored, and the distortion effect caused by unstable light is reduced. By using the image distortion data and the distortion data for image quality instability analysis, the performance of the commodity image under different optical conditions can be comprehensively evaluated, providing an accurate theoretical basis for image processing and optimization, thereby ensuring the efficiency and stability of the commodity recognition system in practical applications.

[0046] Preferably, step S234 includes the following steps:

[0047] The multi-directional dispersion of scattered light is estimated based on the scattering irregularity enhancement data of commodity categories to obtain the multi-directional dispersion data of scattered light;

[0048] Based on the scattered light multi-directional dispersion data, the image concentrated light source acquisition failure detection is performed to obtain the concentrated light source deviation data of the product image;

[0049] Based on the concentrated light source deviation data of the commodity image, the blur growth of the commodity image is predicted to obtain the blur growth data of the commodity image;

[0050] Perform product image fine-grained loss analysis based on product image fuzziness growth data to obtain product image fine-grained loss data;

[0051] The light spot probability of the product image is calculated based on the multi-directional dispersion data of the scattered light to obtain the light spot probability data of the product image;

[0052] The detail loss data of the commodity image is estimated by using the spot probability data of the commodity image and the fine-grained loss data of the commodity image to obtain the detail loss data of the commodity category image.

[0053] The present invention effectively estimates the multi-directional dispersion of scattered light by analyzing the enhanced data of the scattering irregularity of commodity categories, thereby providing an early warning for the light source deviation problem in image acquisition. The use of scattered light multi-directional dispersion data for centralized light source acquisition failure detection can accurately identify image quality problems caused by light source deviation, providing an important basis for subsequent correction operations. On this basis, by analyzing the centralized light source deviation data, the growth trend of the blurriness of commodity images is predicted, thereby helping to foresee and avoid the occurrence of image blur. In addition, the blur growth data can also be used for fine-grained loss analysis, accurately identifying the loss of details in the image, and providing direction for image quality recovery. At the same time, the multi-directional dispersion data of scattered light can be used to calculate the spot probability of commodity images, further quantify the spot problem in the image, thereby strengthening the control of the overall image quality. By combining fine-grained loss data with spot probability data, it is possible to achieve effective prediction of the loss of details in commodity images, ensure the stability and clarity of the image under complex lighting conditions, and greatly improve the robustness and accuracy of the commodity recognition system.

[0054] Preferably, step S3 comprises the following steps:

[0055] Step S31: performing commodity target detection according to the commodity category image contrast enhancement data, thereby obtaining commodity category image target data;

[0056] Step S32: performing commodity category outline recognition according to the commodity category image target data, thereby obtaining commodity category outline data;

[0057] Step S33: performing a commodity category texture feature analysis based on the commodity category target data to obtain commodity category texture feature data;

[0058] Step S34: extracting surface subtle texture gradients based on the product category texture feature data to obtain product surface subtle texture gradient data;

[0059] Step S35: Calculate the commodity category image feature vector according to the commodity surface subtle texture gradient data and the commodity category texture feature data to obtain the commodity category image feature vector data.

[0060] The present invention detects targets based on the contrast enhancement data of commodity category images, accurately locates the commodity area in the image, effectively extracts the specific information of the commodity, and provides a clear target area for subsequent image analysis. This process helps to remove background noise and improve the efficiency and accuracy of image processing. Further, through contour recognition, the outer shape boundary of the commodity is accurately defined, providing the necessary spatial structure data for subsequent morphological analysis. Texture feature analysis deeply explores the details of the commodity surface, such as material changes and surface structure, and reveals the subtle features of the commodity. Combined with the surface subtle texture gradient extraction technology, it is possible to further explore the surface details of the commodity, enhance the layering and fineness of the texture, ensure that every detail of the commodity is fully identified, integrate the texture feature data and the subtle texture gradient information, and convert the visual features of the commodity into digital features that can be used for classification and identification through image feature vector calculation. The generation of this data provides a solid foundation for subsequent commodity classification and identification, and improves the accuracy and intelligence level of the recognition system.

[0061] The present invention also provides a commodity identification system based on an image classification model, which is used to execute the commodity identification method based on an image classification model as described above. The commodity identification system based on an image classification model includes:

[0062] The physical parameter analysis module is used to obtain basic information data of commodities; perform physical parameter analysis of commodity categories according to the basic information data of commodities to obtain physical parameter data of commodity categories;

[0063] The image processing module is used to collect commodity category images according to the commodity category physical parameter data, thereby obtaining commodity category image data; and perform commodity image contrast enhancement processing on the commodity category image data to obtain commodity category image contrast enhanced data;

[0064] A feature vector calculation module is used to perform commodity target detection based on the commodity category image contrast enhancement data, thereby obtaining commodity category image target data; perform commodity category image feature vector calculation based on the commodity category image target data, thereby obtaining commodity category image feature vector data;

[0065] The commodity classification and recognition module is used to analyze the subtle features of the external morphology of the commodity category based on the commodity category image feature vector data to obtain the subtle feature data of the external morphology of the commodity category; adjust the model recognition data of the subtle feature data of the external morphology of the commodity category based on the picture classification model to obtain the picture classification and recognition adjustment model; and perform commodity classification and recognition processing on the basic information data of the commodity based on the picture classification adjustment model to obtain commodity classification and recognition data.

[0066] The present invention also provides a vending machine, including a vending machine main body, a power supply unit and an electrical control unit. The power supply unit is installed inside the vending machine main body, the electrical control unit is electrically connected to the power supply unit, the electrical control unit is used to charge and control the vending machine main body, and the electrical control unit is used to execute a commodity identification method based on an image classification model as described above.

[0067] The present invention is that, by implementing the above method, the accuracy and robustness of the commodity recognition system can be effectively improved, the basic information data of the commodity is obtained and the physical parameter analysis is performed, which can provide a reliable basis for subsequent image acquisition and processing, and ensure that the system can identify and distinguish different categories of commodities. The analysis results of the physical parameter data of the commodity category provide a valuable reference for the image acquisition process, help identify the size, weight and other information of the item, and these information can optimize the direction and method of image acquisition, so that the acquired image can more accurately reflect the appearance characteristics of the commodity. After the commodity image is acquired and processed by contrast enhancement, the image quality problems caused by factors such as ambient lighting and shooting angle are eliminated, and the clarity and details of the image are ensured, thereby effectively avoiding the influence of low-quality images on the subsequent recognition accuracy. Image contrast enhancement can not only improve the detail performance of the image, but also highlight the outline of the commodity, making the target detection more accurate, and providing high-quality data for subsequent image feature extraction. With the completion of commodity target detection, the commodity area in the image can be accurately identified and further processed. The commodity category image target data provides a clear basis for the subsequent image feature vector calculation, so that the unique features of the commodity can be extracted and quantified. The generation of feature vectors provides an objective basis for the identification of goods. These feature data can significantly improve the accuracy of commodity classification and avoid misjudgment caused by factors such as similar appearance or lighting. In combination with the picture classification model, the subtle features of the external morphology of the commodity category are deeply analyzed and the model is adjusted to further optimize the recognition accuracy. Through the adjusted classification model, the basic information data of the commodity is processed, the category of the commodity is accurately identified, and the subtle differences between different commodities are accurately distinguished, which not only improves the accuracy of commodity classification, but also enhances the adaptability of the system to commodity images in complex scenes, and improves the efficiency and accuracy of recognition. Through accurate commodity recognition, the entire system can achieve efficient and reliable automated commodity recognition and classification processing, providing more powerful technical support for applications. Therefore, the present invention is an optimization process made to a traditional commodity recognition method based on a picture classification model, which solves the problem of inaccurate commodity recognition in a traditional commodity recognition method based on a picture classification model, and the problem of inaccurate recognition of abnormal commodity category images, improves the accuracy of commodity recognition, and the accuracy of abnormal recognition of commodity category images. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] Figure 1A schematic diagram of the steps of a commodity identification method based on an image classification model;

[0069] Figure 2 for Figure 1 Detailed implementation steps of step S2 in the flowchart;

[0070] Figure 3 for Figure 1 Detailed implementation steps of step S3 in FIG.

[0071] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0072] The technical method of the present invention is described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by technicians in this field without creative work are within the scope of protection of the present invention.

[0073] In addition, the accompanying drawings are only schematic illustrations of the present invention and are not necessarily drawn to scale. The same reference numerals in the figures represent the same or similar parts, and their repeated description will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities and do not necessarily correspond to physically or logically independent entities. The functional entities can be implemented in software form, or implemented in one or more hardware modules or integrated circuits, or implemented in different networks and / or processor methods and / or microcontroller methods.

[0074] It should be understood that, although the terms "first", "second", etc. may be used herein to describe various units, these units should not be limited by these terms. These terms are used only to distinguish one unit from another unit. For example, without departing from the scope of the exemplary embodiments, the first unit may be referred to as the second unit, and similarly the second unit may be referred to as the first unit. The term "and / or" used herein includes any and all combinations of one or more of the listed associated items.

[0075] To achieve this, please refer to Figures 1 to 3 , a commodity identification method based on an image classification model, comprising the following steps:

[0076] Step S1: Obtain basic information data of the commodity; perform commodity category physical parameter analysis based on the commodity basic information data to obtain commodity category physical parameter data;

[0077] Step S2: collecting commodity category images according to the commodity category physical parameter data, thereby obtaining commodity category image data; performing commodity image contrast enhancement processing on the commodity category image data, thereby obtaining commodity category image contrast enhanced data;

[0078] Step S3: performing commodity target detection according to the commodity category image contrast enhancement data, thereby obtaining commodity category image target data; performing commodity category image feature vector calculation according to the commodity category image target data, thereby obtaining commodity category image feature vector data;

[0079] Step S4: Analyze the subtle features of the external morphology of the commodity category based on the commodity category image feature vector data to obtain the subtle feature data of the external morphology of the commodity category; adjust the model recognition data of the subtle feature data of the external morphology of the commodity category based on the picture classification model to obtain the picture classification recognition adjustment model; perform commodity classification recognition processing on the commodity basic information data based on the picture classification adjustment model to obtain the commodity classification recognition data.

[0080] In the embodiment of the present invention, reference Figure 1 As shown, in this example, the commodity identification method based on the image classification model includes the following steps:

[0081] Step S1: Obtain basic information data of the commodity; perform commodity category physical parameter analysis based on the commodity basic information data to obtain commodity category physical parameter data;

[0082] In an embodiment of the present invention, basic information data of a commodity is obtained, including key data such as the category, brand, material, and size of the commodity. This process is completed by accessing a commodity database or collecting real-time data through a sensor. Then, a physical parameter analysis of the commodity category is performed based on the acquired basic information data. In this analysis, data such as the material and size of the commodity will be input into a physical formula to calculate physical parameters such as the weight, density, and volume of the commodity. The specific analysis steps include: calculating the density of the substance based on the material data, and performing volume calculation in combination with the size data of the commodity, and then inferring the mass of the commodity. Through these calculations, the physical parameter data of the commodity category is obtained.

[0083] Step S2: collecting commodity category images according to the commodity category physical parameter data, thereby obtaining commodity category image data; performing commodity image contrast enhancement processing on the commodity category image data, thereby obtaining commodity category image contrast enhanced data;

[0084] In an embodiment of the present invention, image acquisition of commodity categories is performed based on the physical parameter data of commodity categories. This step obtains high-definition image data of commodities through a high-precision camera or sensor. During the acquisition process, the image acquisition device should have functions such as automatic adjustment of focal length and exposure time to ensure the best image quality. Subsequently, the collected commodity image data is subjected to contrast enhancement processing. The specific operation is to enhance the brightness contrast of the image through an image processing algorithm (such as histogram equalization or adaptive contrast adjustment), so that the edges, details and textures of the commodities are clearer.

[0085] Step S3: performing commodity target detection according to the commodity category image contrast enhancement data, thereby obtaining commodity category image target data; performing commodity category image feature vector calculation according to the commodity category image target data, thereby obtaining commodity category image feature vector data;

[0086] In an embodiment of the present invention, target detection is performed on the commodity image after contrast enhancement processing. The target detection process uses classic computer vision algorithms, such as YOLO (You Only Look Once) or Faster R-CNN, which can quickly identify the bounding box of the commodity in the image and mark the area where the commodity is located. The detected target area contains basic morphological information of the commodity, such as contours, edges, etc. Afterwards, the commodity category image feature vector is calculated based on the detected commodity target area data. In the feature vector calculation process, a convolutional neural network (CNN) is used to extract high-dimensional features of the image, obtain feature information such as texture, shape, color, etc. of the commodity, and convert it into feature vector data in numerical form.

[0087] Step S4: Analyze the subtle features of the external morphology of the commodity category based on the commodity category image feature vector data to obtain the subtle feature data of the external morphology of the commodity category; adjust the model recognition data of the subtle feature data of the external morphology of the commodity category based on the picture classification model to obtain the picture classification recognition adjustment model; perform commodity classification recognition processing on the commodity basic information data based on the picture classification adjustment model to obtain the commodity classification recognition data.

[0088] In an embodiment of the present invention, the subtle features of the external morphology of the commodity are analyzed based on the feature vector data of the commodity category image. In this analysis, texture analysis algorithms, such as gray level co-occurrence matrix (GLCM) and local binary pattern (LBP), are used to extract subtle surface features of the commodity, such as surface texture and detail structure. These features can help distinguish subtle differences between commodities of different categories. Next, the model recognition data is adjusted for the subtle feature data of the external morphology of the commodity category based on the image classification model. In this process, the feature data of the commodity is trained and optimized using the existing commodity classification model, and the parameters of the model are adjusted to improve the classification accuracy. Based on the adjusted image classification model, the commodity basic information data is processed for commodity classification recognition to obtain commodity classification recognition data. After this step is completed, the system can accurately identify the category of the commodity, such as electronic products, daily commodities, etc., and give the corresponding classification results.

[0089] Preferably, step S1 comprises the following steps:

[0090] Step S11: Obtain basic information data of the product;

[0091] Step S12: collecting commodity category and brand data according to commodity basic information data, thereby obtaining commodity category and brand data;

[0092] Step S13: performing commodity category material analysis according to commodity category brand data, thereby obtaining commodity category material data;

[0093] Step S14: measuring the commodity category size according to the commodity category brand data, thereby obtaining the commodity category size data;

[0094] Step S15: Perform commodity category physical parameter analysis based on commodity category size data and commodity category material data to obtain commodity category physical parameter data.

[0095] In an embodiment of the present invention, the acquisition of commodity basic information data involves collecting a number of basic information about the commodity, including commodity name, brand, model, manufacturer, material, size, etc. The data is obtained in a variety of ways, for example, using barcode scanning technology to extract information from commodity packaging, or automatically obtaining from a warehouse database through a supply chain management system. The key to this step is to ensure that the information obtained is accurate and complete, so as to provide the necessary basis for subsequent analysis. Common tools include barcode scanners, database interface tools, and manual data input interfaces. Through the brand field in the commodity basic information data, the brand category to which the commodity belongs is further extracted. For commodities with existing brands, the brand information is matched with the existing commodity brand data in the database through an automated system. This process often relies on a database search function, which can classify commodities according to the basic information fields of the commodity (such as the brand field). For unknown brand commodities, brand information is obtained from suppliers or e-commerce platforms through data crawling technology, or collected by manually checking brand logos. After obtaining the brand information of the commodity, material analysis is performed according to the commodity category to which the brand belongs, and the commodity brand will be associated with a specific type of material (for example, a certain brand of mobile phone usually uses aluminum alloy material, and a certain brand of furniture uses solid wood or board material). According to the brand data of the product, the material data is analyzed by querying the brand-related material library or directly through the product manual provided by the manufacturer. The analysis is achieved through automated database matching, or in some cases, the material is inferred by manually comparing the physical properties of the product. The size data of the product is usually obtained through the product label, product manual or product details page on the e-commerce platform. According to the brand and category of the product, the product size data is automatically obtained using the relevant product database or the open API of the e-commerce platform. If the size data is missing or not directly available, the measurement tool is used for physical measurement. For example, for clothing products, rulers and measurement software are used to accurately record the length, width, height and other size information of the product. For more complex items (such as furniture, electronic equipment, etc.), scanning equipment is used to perform three-dimensional measurement and generate relevant size data. The size data usually includes length, width, height, diameter, etc. After obtaining the size data and material data of the product, the physical parameter data of the product is obtained through analysis by physical formulas, and the volume of the product is calculated based on the size data of the product. This is usually performed through three-dimensional measurement (length × width × height) or applicable geometric calculation methods (such as cylinders, spheres, etc.). Then, according to the characteristics of the product material, such as density, elastic modulus, etc., the material data is further calculated to obtain the physical parameters of the product such as mass, weight, density, etc. If the material is a composite material, the overall density is estimated by the known composite material composition ratio. The result of this step is a series of physical parameter data, such as the weight, volume, density, mass, etc. of the product, which is further processed by image processing and target detection.

[0096] Preferably, step S15 comprises the following steps:

[0097] Step S151: Calculate the volume of the commodity category according to the commodity category size data, thereby obtaining the commodity category volume data;

[0098] Step S152: collecting commodity category density according to commodity category material data, thereby obtaining commodity category density data;

[0099] Step S153: Calculate the commodity category quality according to the commodity category quality measurement data and the commodity category volume data, thereby obtaining the commodity category quality data;

[0100] Step S154: Calculate the commodity weight according to the commodity category mass data and the commodity category volume data to obtain the commodity category weight data;

[0101] Step S155: Detecting the surface roughness of the commodity according to the commodity category material data, thereby obtaining the commodity category surface roughness data;

[0102] Step S156: Perform commodity category physical parameter analysis based on commodity category surface roughness data and commodity category weight data to obtain commodity category physical parameter data.

[0103] In an embodiment of the present invention, the geometric shape of the commodity is determined according to the size data of the commodity. For example, for a rectangular commodity, the volume is calculated by the formula: volume = length × width × height; for a cylindrical commodity, the formula is used: volume = π × radius 2 × height. For irregularly shaped commodities, 3D scanning technology is used to obtain its three-dimensional data, and the scanning data is used to calculate its volume. These volume data provide a basis for subsequent mass and weight calculations. If the commodity is a liquid or granular item, the volume is estimated by volume measurement or comparison with a known standard. The tools used include a laser rangefinder, a 3D scanner, or image-based dimension measurement software. According to the material data of the commodity, the density data usually comes from the technical specifications of the product or the material property data already in the database. For common materials, such as metals, plastics, wood, etc., their density has a clear value in the standard database. If the material is a mixture or a composite material, a weighted average calculation is performed by the known component ratio and the density of each component. In addition, for some special commodities, such as liquid commodities, the density parameters are collected through experiments or provided. In order to improve the accuracy of the data, an electronic densitometer or a density measuring instrument is used for actual measurement. The output of this step is the density data of the product category, and the common unit is kilograms / cubic meter. After obtaining the volume and density data of the product, according to the physical formula, the mass of the product is calculated by the formula: mass = density × volume. If the product is irregular in shape, it is still calculated by the acquired three-dimensional volume data and the material density value. If the product is a liquid or an item with special physical properties, the mass value is obtained through dynamic measurement or reference data. In mass measurement, the tools used include density meters, mass sensors, or estimation through image analysis combined with a database. This step obtains the mass data of the product, and the common units are kilograms (kg) or grams (g). The weight of the product is calculated based on its mass and gravitational acceleration. Using the known mass data of the product, combined with the standard gravitational acceleration (9.8m / s 2) to calculate the weight, the formula is: weight = mass × gravitational acceleration. For most products, the weight data is usually calculated directly by the above formula. In some application scenarios, the product is in different gravity environments, and the corresponding acceleration value needs to be used to adjust the calculation. This step obtains the weight data of the product, and the common unit is kilogram (kg) or gram (g). The surface roughness of the product is an important indicator of the appearance quality and performance of the product. According to the material data of the product, select an appropriate surface roughness detection method. For metal products, surface roughness detection can be performed using equipment such as a profilometer or a laser scanner. For products made of materials such as plastic and wood, a stylus roughness meter or three-dimensional laser scanning technology is used to measure the surface roughness. The roughness value is usually expressed in units of Ra (arithmetic mean roughness), which reflects the smoothness of the undulations on the surface of the product. If the product is a liquid or colloidal substance, its surface roughness can be indirectly inferred by surface tension measurement or related physical methods. After obtaining the surface roughness data and weight data of the product, the physical parameters of the product are analyzed. Roughness is related to the friction characteristics and wear resistance of the product, while weight is related to factors such as the gravity and carrying capacity of the product. Through the comprehensive analysis of these parameters, more physical properties of the product, such as strength and durability, can be obtained. For example, products with higher surface roughness will show greater friction and are suitable for certain usage scenarios. The analysis is performed through manual evaluation or comprehensive calculation through existing physical models (such as object surface characteristic analysis models) to obtain physical parameter data.

[0104] Preferably, step S2 comprises the following steps:

[0105] Step S21: collecting commodity category images according to commodity category physical parameter data, thereby obtaining commodity category image data;

[0106] Step S22: performing commodity image preprocessing on the commodity category image data, thereby obtaining commodity category image preprocessing data;

[0107] Step S23: performing commodity category image anomaly detection based on commodity category image preprocessing data and commodity category physical parameter data to obtain commodity category image anomaly data;

[0108] Step S24: performing product image contrast enhancement processing on the product category image abnormal data to obtain product category image contrast enhanced data.

[0109] As an example of the present invention, refer to Figure 2 As shown, in this example, step S2 includes:

[0110] Step S21: collecting commodity category images according to commodity category physical parameter data, thereby obtaining commodity category image data;

[0111] In an embodiment of the present invention, the conditions and equipment selection for collecting commodity images are determined based on the physical parameter data of the commodity. For different physical properties of commodities such as size, weight and material, appropriate image acquisition equipment and angles are selected. Image acquisition equipment includes high-definition cameras, digital cameras, scanners or 3D scanners, etc., and it is necessary to ensure that the commodity is in a stable environment during acquisition to avoid image deviation caused by factors such as lighting and angle. If the commodity is a small object, macro photography equipment can be used; for large commodities, professional studios or panoramic photography technology can be used for acquisition. Image acquisition should take into account resolution and frame rate to ensure the clarity of the image. The resolution is usually selected to be no less than 300DPI to ensure that details are not lost in subsequent processing and to obtain the image data of the commodity.

[0112] Step S22: performing commodity image preprocessing on the commodity category image data, thereby obtaining commodity category image preprocessing data;

[0113] In the embodiment of the present invention, after the product image data is collected, image preprocessing is required to improve the accuracy of subsequent analysis and recognition, and image denoising is performed to remove noise in the image by methods such as median filtering, mean filtering or Gaussian filtering to ensure that the image quality is not affected by external interference. Secondly, the image is color corrected, and the color deviation in the image is adjusted using a color balance algorithm so that the true color of the product is accurately presented. Furthermore, according to the angle of image acquisition, the image is geometrically corrected, and image transformation techniques (such as perspective transformation, rotation, cropping, etc.) are used to ensure that the display effect of the product is consistent with actual observations. The image is edge detected, such as the Canny edge detection algorithm, to extract the contour of the product. The image data obtained after image preprocessing should have high clarity and accuracy.

[0114] Step S23: performing commodity category image anomaly detection based on commodity category image preprocessing data and commodity category physical parameter data to obtain commodity category image anomaly data;

[0115] In an embodiment of the present invention, after image preprocessing, it is necessary to perform anomaly detection on the product image to identify existing quality problems or inconsistent image content. Anomaly detection is achieved by performing pattern recognition and comparison analysis on the image preprocessing data, combining the physical parameter data of the product (such as size, weight, surface material, etc.) with the image data, and determining whether there is an anomaly by calculating the contrast, texture, shape and other features of each part of the product image. For example, if a part of the product image is distorted or abnormally illuminated, it will affect the subsequent recognition results. Computer vision technology is used to analyze the image using image matching algorithms, shape detection algorithms, and target recognition algorithms to detect whether there are image areas that do not meet expectations. If an abnormal part is detected, it will be marked as image abnormality data. Common algorithms include convolutional neural network (CNN) anomaly detection models based on deep learning, or SVM, K-means clustering and other methods based on traditional machine learning.

[0116] Step S24: performing product image contrast enhancement processing on the product category image abnormal data to obtain product category image contrast enhanced data.

[0117] In an embodiment of the present invention, after detecting an abnormality in a commodity category image, the image needs to be contrast enhanced to improve image quality and detail display. Contrast enhancement ensures that the details of the commodity are more obvious by adjusting the brightness and color saturation of different regions in the image. Commonly used methods include histogram equalization and adaptive histogram equalization (CLAHE), which can enhance local details of the image, especially in areas with large brightness changes. For the processing of abnormal areas, a local contrast enhancement method is used to restore the details of the image by enhancing the contrast of a specific area of ​​the image to make it more consistent with physical reality. Edge enhancement methods such as the Laplace operator are applied to further highlight the edges and details of the commodity, so as to facilitate subsequent target detection and classification recognition.

[0118] Preferably, step S1 comprises the following steps:

[0119] Step S231: Calculate the surface reflectivity of the commodity category according to the physical parameter data of the commodity category to obtain the surface optical refractive index data of the commodity category;

[0120] Step S232: performing commodity image quality instability analysis on the commodity category image preprocessing data according to the commodity category surface optical refractive index data to obtain commodity image quality instability analysis;

[0121] Step S233: performing commodity category scattering irregularity enhancement analysis according to the commodity category physical parameter data to obtain commodity category scattering irregularity enhancement data;

[0122] Step S234: performing detail loss estimation on the commodity category image preprocessing data according to the commodity category scattering irregularity enhancement data to obtain commodity category image detail loss data;

[0123] Step S235: performing commodity category image anomaly detection based on commodity category image detail loss data and commodity image quality instability analysis to obtain commodity category image anomaly data.

[0124] In the embodiment of the present invention, according to the physical parameter data of the commodity category, the reflectivity of the commodity surface needs to be calculated by the known surface material and optical properties. Reflectivity is the ratio of light reflected by the surface of an object, which directly affects the gloss and detail presentation of the image. In this step, the Fresnel equation is used to calculate the reflectivity by collecting the material information of the commodity surface, such as the material type (for example, metal, glass, plastic, etc.) and physical properties such as surface roughness, combined with the incident light angle and the surface normal. For transparent or translucent objects, the refraction and transmission of light must also be considered to calculate the optical refractive index of the object. The optical refractive index data of the surface of the commodity is calculated using the law of optical refraction. The refractive index data is crucial to the subsequent image quality analysis and will affect the propagation mode and visual effect of light in the image. The stability analysis of the commodity image quality is performed by the optical refractive index data of the commodity category and the pre-processed image data. By calculating the reflected light intensity and refracted light changes in different areas of the image, combined with the optical refractive index data of the commodity surface, the performance of the commodity image under different lighting conditions is analyzed. If the image has large fluctuations at different angles or lighting, problems such as light spots, overexposure or shadows will occur, thereby affecting the stability of the image quality. To address these issues, the degree of instability of the product image quality is obtained by analyzing the image's color deviation, brightness fluctuation and other characteristics. This analysis is performed by calculating the local and global contrast of the image, the standard deviation of the brightness value, and the high-frequency noise that appears in the image. The results of this step provide evaluation data for the product image quality, reflecting the impact of the optical refractive index on the image performance. The surface material of the product has an important influence on the scattering characteristics of light, especially under complex lighting conditions. According to the physical parameter data of the product (such as the granularity, surface roughness, and light transmittance of the material), its optical scattering is enhanced and analyzed. A scattering model, such as the Monte Carlo ray tracing method, is used to simulate the interaction between the light on the product surface and the medium, and the scattering angle and intensity distribution of light are calculated. By comparing and analyzing the scattering laws of different materials, the irregular scattering phenomenon is enhanced, especially the reflection or refraction changes shown on the product surface under uneven lighting. The purpose of the scattering irregularity enhancement analysis is to improve the details and realism of the image, and to show the irregular optical characteristics of the product surface by strengthening the scattering effect. The processed data helps to enhance the layering and realism of the product in the image. According to the scattering irregularity enhancement data of the product category, the detail loss of the preprocessed image is estimated. When some details in an image are lost or blurred due to excessive scattering or uneven illumination, the degree of detail loss in the image is estimated based on the scattering irregularity enhancement data. By comparing the enhanced scattering effect with the original image, the impact of factors such as uneven illumination and uneven surface of the object on the image details is analyzed. For areas where details are lost, methods such as Gaussian blur, gradient change, and texture information analysis are used to evaluate the degree of image loss.The estimated results will be provided in the form of data, indicating the severity of the loss of image details, especially when the reflection or scattering effect of the product surface is large, the loss of details is more obvious, affecting the identification and classification of the product. Based on the detail loss estimation data in step S234 and the image quality instability analysis results, the product image anomaly detection is performed. This step combines the degree of detail loss and quality instability of the image and adopts a multi-dimensional anomaly detection method for analysis. Specifically, by calculating the texture changes, brightness fluctuations, contrast anomalies and other features of the image, combined with the degree of detail loss, it is judged whether the image has abnormal problems. Commonly used anomaly detection methods include the standard deviation method based on statistical analysis, or the support vector machine (SVM) and clustering algorithm based on machine learning. The goal of this step is to identify and mark abnormal image data caused by changes in ambient light and surface material characteristics during image acquisition.

[0125] Preferably, step S232 includes the following steps:

[0126] According to the optical refractive index data of the commodity category surface, the commodity category refractive index is divided into the commodity category surface high refractive index data and the commodity category surface low refractive index data;

[0127] According to the high refractive index data of the commodity category surface, the change of the light propagation direction of the commodity surface is estimated to obtain the change data of the light propagation direction of the commodity surface;

[0128] The product category image distortion is estimated based on the data of the change in the light propagation direction on the product surface, and the product category image distortion data is obtained;

[0129] Based on the low refractive index data of the commodity category surface, the excessive reflection of the light of the commodity category is estimated to obtain the excessive reflection data of the light of the commodity category;

[0130] According to the product category light over-reflection data and the product category image distortion data, the product category image preprocessing data is estimated to obtain the product category image distortion data;

[0131] The commodity category image distortion data and the commodity category image deformation data are used to perform commodity image quality instability analysis to obtain commodity image quality instability analysis.

[0132] In the embodiment of the present invention, the optical refractive index data of the commodity category is an important parameter for describing the refractive characteristics of light on the commodity surface. By analyzing the surface material of the commodity and its corresponding refractive index, the commodity is divided into two categories: high refractive index and low refractive index. In the specific implementation process, the physical property data of different commodity categories are collected to measure their surface refractive index. The refractive index formula (such as Snell's law) is used for calculation to obtain the refractive index value of each commodity surface, and the commodities are classified into two categories: high refractive index and low refractive index based on the preset refractive index range. Commodities with high refractive index are usually materials with smooth surfaces and strong light reflection characteristics such as glass and metal, while low refractive index commodities are plastics or some materials with strong light absorption. Through this step, the optical characteristics of the commodities are effectively distinguished, thereby providing basic data for subsequent optical behavior prediction. According to the high refractive index data of the commodity category surface, the propagation direction of light on the surfaces of these commodities is predicted. The high refractive index surface of the commodity has a strong light refraction ability, and the light will refract and change its propagation path after entering these surfaces. In this step, an optical simulation model (such as ray tracing technology) is used to estimate the light propagation direction of the high refractive index surface of the product. By inputting the refractive index of the product surface, the incident angle of light, and the surface normal, the propagation direction of the refracted light is calculated. For each product category, the propagation trajectory of light on the product surface is accurately predicted based on the refractive index and related physical properties of its surface material, and the data on the change in light propagation direction on the product surface is obtained. The image distortion of the product category is estimated through the data on the change in light propagation direction on the product surface. When light passes through a high refractive index surface, the image is distorted due to the change in the refraction angle. For example, the object appears to be stretched, compressed, or twisted. This step simulates the refraction effect on the product surface and estimates the distortion of the image in combination with the geometric shape of the product. Use optical distortion models, such as barrel distortion and pincushion distortion models, to analyze the product image, and estimate the degree of image deformation in combination with the change data on the light propagation direction on the product surface. In this process, the key data input include factors such as the surface refractive index of the product, the propagation path of light, and the angle of the image. The image distortion data obtained is used to estimate the excessive reflection of light by simulating the propagation and reflection characteristics of light for the low refractive index data of the product category. The surface of low-refractive-index products usually has strong light reflection, which causes the image to be too bright, details to be lost, or the highlight area cannot be accurately restored. In this step, by combining parameters such as the refractive index, surface roughness, and light intensity of the product surface, the surface reflection characteristics of the product are calculated using an optical reflection model (such as the Fresnel equation) to predict the intensity and angle of light reflected on the product surface. Based on the prediction results, the degree of excessive reflection of light on the product surface is obtained, and the excessive reflection data of light of the product category is obtained. Combining the excessive reflection data of light of the product category with the image distortion data, the distortion of the product category image is estimated.Excessive reflection and distortion are important factors affecting image quality. Excessive reflection of light can make the image too bright or overexposed, while image distortion can cause the geometric shape of the image to change. This step estimates the degree of distortion of the commodity category image by analyzing the reflection intensity and refraction of light and combining distortion models (such as perspective distortion or nonlinear deformation). By comparing the difference between the preprocessed image and the ideal image, the distortion caused by excessive light reflection and surface distortion is detected, and the commodity category image distortion data is obtained. Based on the obtained commodity category image distortion data and image distortion data, the instability of commodity image quality is analyzed. The instability of image quality is mainly manifested in uneven image brightness, too high or too low contrast, and loss of details. By combining the degree of image distortion and distortion, the image quality is comprehensively evaluated using image quality evaluation indicators (such as image contrast, color difference, brightness uniformity, etc.). In this process, based on the global and local features of the image, the fluctuation range of image quality is calculated, and the stability of the commodity image is quantified.

[0133] Preferably, step S234 includes the following steps:

[0134] The multi-directional dispersion of scattered light is estimated based on the scattering irregularity enhancement data of commodity categories to obtain the multi-directional dispersion data of scattered light;

[0135] Based on the scattered light multi-directional dispersion data, the image concentrated light source acquisition failure detection is performed to obtain the concentrated light source deviation data of the product image;

[0136] Based on the concentrated light source deviation data of the commodity image, the blur growth of the commodity image is predicted to obtain the blur growth data of the commodity image;

[0137] Perform product image fine-grained loss analysis based on product image fuzziness growth data to obtain product image fine-grained loss data;

[0138] The light spot probability of the product image is calculated based on the multi-directional dispersion data of the scattered light to obtain the light spot probability data of the product image;

[0139] The detail loss data of the commodity image is estimated by using the spot probability data of the commodity image and the fine-grained loss data of the commodity image to obtain the detail loss data of the commodity category image.

[0140] In an embodiment of the present invention, the scattering irregularity enhancement data of the commodity category is analyzed in detail by describing the scattering behavior of the commodity surface at different angles to the light. In this step, the light scattering characteristics of the commodity surface are collected, especially the influence of the surface microstructure (such as texture, roughness, etc.) on the reflection, refraction and scattering of light. Then, a scattering model (such as Mie scattering theory) is used to predict the scattering path of light on the commodity surface. In this process, the material and morphology of the commodity surface are taken into account, the scattering direction of light is simulated by statistical methods, the multi-directional distribution of scattered light is calculated, and the multi-directional dispersion data of scattered light is obtained by integrating the scattering angle and scattering intensity of light in different directions. The obtained multi-directional dispersion data of scattered light is used to detect whether the collection of the concentrated light source in the commodity image is successful. The deviation of the light source in the commodity image is usually manifested as uneven image brightness or uneven distribution of local light spots, especially when the scattered light on the commodity surface is relatively strong. In this step, by analyzing the multi-directional distribution of scattered light, combined with the ideal position of the concentrated light source in the image, the offset of the light source in the actual image acquisition process is identified. Use image processing algorithms (such as Fourier transform or edge detection) to analyze the light spots in the product image to detect whether there is light source deviation. By comparing with the ideal light source distribution predicted by the theoretical model, the degree of light source deviation is calculated, and the light source deviation data in the product image is generated. The blurriness of the product image is usually related to the distribution of the light source and the matching degree of the focal length. The concentrated light source deviation of the image will directly lead to inaccurate image focus, thereby affecting the clarity of the image. In this step, based on the concentrated light source deviation data of the product image, the optical model is used to predict the blur growth of the image, the focusing problem caused by light source deviation in the product image is analyzed, and the image sharpness index (such as Laplace transform or gradient algorithm) is used to calculate the clarity change of the image. Then, combined with the light source deviation data, the blur model is used to estimate the degree of influence of light source deviation on image blurriness, and the blur growth data of the product image is obtained. The blur growth of the product image is usually accompanied by the loss of details in the image, especially in the boundary performance of subtle textures and small objects. In this step, the loss of details in the image due to blur is further analyzed through the blur growth data of the product image, and the high-frequency details in the product image are extracted using an image detail extraction algorithm (such as wavelet transform or edge detection algorithm). Next, the image blur growth data is compared with the detail loss to evaluate the specific impact of blur on the details. The fine-grained loss data of the product image is obtained by calculating the changes in the details in the image pixel by pixel. The scattering characteristics of the light on the surface of the product directly affect the distribution of the light spots in the image, especially on the surfaces of products with high reflection or refraction. The probability of the light spots appearing is calculated by scattering the data of scattered light in multiple directions.In this step, the probability distribution of the light spot in the product image is calculated based on the multi-directional scattered light data of the product category using a probability statistical model (such as Gaussian distribution or Poisson distribution). By analyzing the light spot formation mechanism under different scattering angles and intensities, the probability of light spots appearing in each area is obtained. This data will help identify the impact of light spots on image quality during image processing, especially in areas where the light spots have a greater impact, which require special processing or repair. Light spots and detail loss in product images are usually the main factors leading to unstable image quality. In this step, the light spot probability data of the product image and the fine-grained loss data of the product image are combined to estimate the detail loss, and the total degree of detail loss is calculated by the weighted average method based on the probability distribution of the light spots and the detail loss of the image. For areas with a greater impact of light spots, the impact of detail loss is further analyzed by local processing methods (such as local contrast enhancement or regional blur processing). In this process, the image quality assessment model is used to calculate the contribution of light spots to the loss of image details, and the detail loss data of product category images is obtained. This data can effectively guide the image restoration algorithm, restore the details of the image in a targeted manner, and improve the image quality.

[0141] Preferably, step S3 comprises the following steps:

[0142] Step S31: performing commodity target detection according to the commodity category image contrast enhancement data, thereby obtaining commodity category image target data;

[0143] Step S32: performing commodity category outline recognition according to the commodity category image target data, thereby obtaining commodity category outline data;

[0144] Step S33: performing a commodity category texture feature analysis based on the commodity category target data to obtain commodity category texture feature data;

[0145] Step S34: extracting surface subtle texture gradients based on the product category texture feature data to obtain product surface subtle texture gradient data;

[0146] Step S35: Calculate the commodity category image feature vector according to the commodity surface subtle texture gradient data and the commodity category texture feature data to obtain the commodity category image feature vector data.

[0147] As an example of the present invention, refer to Figure 3 As shown, in this example, step S3 includes:

[0148] Step S31: performing commodity target detection according to the commodity category image contrast enhancement data, thereby obtaining commodity category image target data;

[0149] In an embodiment of the present invention, the contrast enhancement data of the commodity image optimizes the brightness and color difference in the image through an image processing algorithm, so that the target area in the commodity image is more obvious. In this step, an image contrast enhancement algorithm, such as histogram equalization or local contrast enhancement, is applied to optimize the visual effect of the commodity image and make the target part of the commodity more prominent. Then, the target area in the commodity image is determined by combining the enhanced image with a classical target detection method (such as edge detection, morphological transformation, Canny edge detection, etc.). These target areas are usually specific parts or edge parts of the commodity. By performing boundary recognition and region division on the target area, the commodity category image target data is obtained.

[0150] Step S32: performing commodity category outline recognition according to the commodity category image target data, thereby obtaining commodity category outline data;

[0151] In the embodiment of the present invention, the commodity category image target data includes the position and boundary information of the commodity in the image. In order to further extract the geometric features of the commodity, contour recognition technology is used, and the commodity target area is processed using an image segmentation method (such as edge detection or threshold segmentation) to extract the contour line of the object. In this process, the Canny edge detection algorithm is used to locate the contour by using the sharp change of the pixel gray value in the image. Then, a contour tracking algorithm (such as the Freeman chain code method or the contour tracking algorithm) is applied to track and record the outer boundary of the target area. This process can not only extract the shape features of the commodity, but also retain the detailed data of the outer contour of the commodity, thereby obtaining the commodity category contour data.

[0152] Step S33: performing a commodity category texture feature analysis based on the commodity category target data to obtain commodity category texture feature data;

[0153] In an embodiment of the present invention, the commodity category target data includes the regional information of the commodity in the image. Next, the texture feature analysis of the commodity is performed based on these regions, and the texture features of the target area in the image are calculated using the gray level co-occurrence matrix (GLCM) algorithm, including indicators such as energy, contrast, and correlation. By analyzing the grayscale relationship between pixels inside the target area, the delicate texture features of the commodity surface can be extracted. Then, the micro-texture structure of the commodity surface is further extracted using methods such as the local binary pattern (LBP). According to the statistical analysis of different texture features, information such as the regularity, roughness, and directionality of the texture is extracted. These features are significantly different between different commodities, and can provide a reliable texture basis for the accurate identification of commodities, thereby obtaining the commodity category texture feature data.

[0154] Step S34: extracting surface subtle texture gradients based on the product category texture feature data to obtain product surface subtle texture gradient data;

[0155] In the embodiment of the present invention, the extraction of subtle texture gradients on the surface of a commodity is to quantify the more detailed texture structure of the commodity surface. According to the texture feature data of the commodity category, by calculating the texture gradient of the target area in the image, the subtle texture changes on the surface of the commodity are extracted, and the image gradient calculation method (such as the Sobel operator or the Prewitt operator) is used to perform pixel-level gradient calculation on the target area of ​​the commodity to identify the texture change rate of the commodity surface in different areas. By performing gradient processing on the grayscale image in the horizontal, vertical and diagonal directions, the gradient value of each pixel point is obtained, and these gradient values ​​reflect the degree of change of the details on the surface of the commodity. Further, by filtering and optimizing the subtle texture gradient, the influence of noise is removed, and the subtle texture gradient data of the commodity surface is obtained.

[0156] Step S35: Calculate the commodity category image feature vector according to the commodity surface subtle texture gradient data and the commodity category texture feature data to obtain the commodity category image feature vector data.

[0157] In an embodiment of the present invention, the calculation of the feature vector of the commodity category image involves combining the subtle texture gradient information of the commodity surface with the overall texture features of the commodity to form a multi-dimensional feature description vector, and performing feature fusion on the subtle texture gradient data of the commodity surface and the commodity category texture feature data. This process uses a feature fusion algorithm (such as principal component analysis PCA or linear discriminant analysis LDA) to reduce the dimension of different texture features and extract the most representative features. In this way, complex texture information is converted into a concise and efficient feature vector, reflecting the most recognizable texture features in the commodity image, and the obtained commodity category image feature vector data will contain comprehensive information on the texture, morphology, etc. of the commodity.

[0158] The present invention also provides a commodity identification system based on an image classification model, which is used to execute the commodity identification method based on an image classification model as described above. The commodity identification system based on an image classification model includes:

[0159] The physical parameter analysis module is used to obtain basic information data of commodities; perform physical parameter analysis of commodity categories according to the basic information data of commodities to obtain physical parameter data of commodity categories;

[0160] The image processing module is used to collect commodity category images according to the commodity category physical parameter data, thereby obtaining commodity category image data; and perform commodity image contrast enhancement processing on the commodity category image data to obtain commodity category image contrast enhanced data;

[0161] A feature vector calculation module is used to perform commodity target detection based on the commodity category image contrast enhancement data, thereby obtaining commodity category image target data; perform commodity category image feature vector calculation based on the commodity category image target data, thereby obtaining commodity category image feature vector data;

[0162] The commodity classification and recognition module is used to analyze the subtle features of the external morphology of the commodity category based on the commodity category image feature vector data to obtain the subtle feature data of the external morphology of the commodity category; adjust the model recognition data of the subtle feature data of the external morphology of the commodity category based on the picture classification model to obtain the picture classification and recognition adjustment model; and perform commodity classification and recognition processing on the basic information data of the commodity based on the picture classification adjustment model to obtain commodity classification and recognition data.

[0163] The present invention also provides a vending machine, including a vending machine main body, a power supply unit and an electrical control unit. The power supply unit is installed inside the vending machine main body, the electrical control unit is electrically connected to the power supply unit, the electrical control unit is used to charge and control the vending machine main body, and the electrical control unit is used to execute a commodity identification method based on an image classification model as described above.

[0164] The above description is only a specific embodiment of the present invention, so that those skilled in the art can understand or implement the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but should conform to the widest scope consistent with the principles and novel features invented herein.

Claims

1. A commodity identification method based on an image classification model, characterized in that: The following steps are involved: Step S1: Obtain basic information data of the commodity; perform commodity category physical parameter analysis based on the commodity basic information data to obtain commodity category physical parameter data; Step S2: collecting commodity category images according to the commodity category physical parameter data, thereby obtaining commodity category image data; Performing product image contrast enhancement processing on the product category image data to obtain product category image contrast enhanced data; Step S3: performing commodity target detection according to the commodity category image contrast enhancement data, thereby obtaining commodity category image target data; Calculate the commodity category image feature vector according to the commodity category image target data to obtain the commodity category image feature vector data; Step S4: Analyze the subtle features of the external morphology of the commodity category based on the commodity category image feature vector data to obtain the subtle feature data of the external morphology of the commodity category; adjust the model recognition data of the subtle feature data of the external morphology of the commodity category based on the picture classification model to obtain the picture classification recognition adjustment model; perform commodity classification recognition processing on the commodity basic information data based on the picture classification adjustment model to obtain the commodity classification recognition data.

2. The commodity identification method based on the image classification model according to claim 1 is characterized in that: Step S1 includes the following steps: Step S11: Obtain basic information data of the product; Step S12: collecting commodity category and brand data according to commodity basic information data, thereby obtaining commodity category and brand data; Step S13: performing commodity category material analysis according to commodity category brand data, thereby obtaining commodity category material data; Step S14: measuring the commodity category size according to the commodity category brand data, thereby obtaining the commodity category size data; Step S15: Perform commodity category physical parameter analysis based on commodity category size data and commodity category material data to obtain commodity category physical parameter data.

3. The commodity identification method based on the image classification model according to claim 2 is characterized in that: Step S1 includes the following steps: Step S151: Calculate the volume of the commodity category according to the commodity category size data, thereby obtaining the commodity category volume data; Step S152: collecting commodity category density according to commodity category material data, thereby obtaining commodity category density data; Step S153: Calculate the commodity category quality according to the commodity category quality measurement data and the commodity category volume data, thereby obtaining the commodity category quality data; Step S154: Calculate the commodity weight according to the commodity category mass data and the commodity category volume data to obtain the commodity category weight data; Step S155: Detecting the surface roughness of the commodity according to the commodity category material data, thereby obtaining the commodity category surface roughness data; Step S156: Perform commodity category physical parameter analysis based on commodity category surface roughness data and commodity category weight data to obtain commodity category physical parameter data.

4. The commodity identification method based on the image classification model according to claim 1, characterized in that: Step S2 includes the following steps: Step S21: collecting commodity category images according to commodity category physical parameter data, thereby obtaining commodity category image data; Step S22: performing commodity image preprocessing on the commodity category image data, thereby obtaining commodity category image preprocessing data; Step S23: performing commodity category image anomaly detection based on commodity category image preprocessing data and commodity category physical parameter data to obtain commodity category image anomaly data; Step S24: performing product image contrast enhancement processing on the product category image abnormal data to obtain product category image contrast enhanced data.

5. The commodity identification method based on the image classification model according to claim 4 is characterized in that: Step S23 includes the following steps: Step S231: Calculate the surface reflectivity of the commodity category according to the physical parameter data of the commodity category to obtain the surface optical refractive index data of the commodity category; Step S232: performing commodity image quality instability analysis on the commodity category image preprocessing data according to the commodity category surface optical refractive index data to obtain commodity image quality instability analysis; Step S233: performing commodity category scattering irregularity enhancement analysis according to the commodity category physical parameter data to obtain commodity category scattering irregularity enhancement data; Step S234: performing detail loss estimation on the commodity category image preprocessing data according to the commodity category scattering irregularity enhancement data to obtain commodity category image detail loss data; Step S235: performing commodity category image anomaly detection based on commodity category image detail loss data and commodity image quality instability analysis to obtain commodity category image anomaly data.

6. The commodity identification method based on the image classification model according to claim 5 is characterized in that: Step S232 includes the following steps: According to the optical refractive index data of the commodity category surface, the commodity category refractive index is divided into the commodity category surface high refractive index data and the commodity category surface low refractive index data; According to the high refractive index data of the commodity category surface, the change of the light propagation direction of the commodity surface is estimated to obtain the change data of the light propagation direction of the commodity surface; The product category image distortion is estimated based on the data of the change in the light propagation direction on the product surface, and the product category image distortion data is obtained; Based on the low refractive index data of the commodity category surface, the excessive reflection of the light of the commodity category is estimated to obtain the excessive reflection data of the light of the commodity category; According to the product category light over-reflection data and the product category image distortion data, the product category image preprocessing data is estimated to obtain the product category image distortion data; The commodity category image distortion data and the commodity category image deformation data are used to perform commodity image quality instability analysis to obtain commodity image quality instability analysis.

7. The commodity identification method based on the image classification model according to claim 5, characterized in that: Step S234 includes the following steps: The multi-directional dispersion of scattered light is estimated based on the scattering irregularity enhancement data of commodity categories to obtain the multi-directional dispersion data of scattered light; Based on the scattered light multi-directional dispersion data, the image concentrated light source acquisition failure detection is performed to obtain the concentrated light source deviation data of the product image; Based on the concentrated light source deviation data of the commodity image, the blur growth of the commodity image is predicted to obtain the blur growth data of the commodity image; Perform fine-grained loss analysis on product images based on the fuzziness growth data of product images to obtain fine-grained loss data of product images; The light spot probability of the product image is calculated based on the multi-directional dispersion data of the scattered light to obtain the light spot probability data of the product image; The detail loss data of the commodity image is estimated by using the spot probability data of the commodity image and the fine-grained loss data of the commodity image to obtain the detail loss data of the commodity category image.

8. The commodity identification method based on the image classification model according to claim 1, characterized in that: Step S3 includes the following steps: Step S31: performing commodity target detection according to the commodity category image contrast enhancement data, thereby obtaining commodity category image target data; Step S32: performing commodity category outline recognition according to the commodity category image target data, thereby obtaining commodity category outline data; Step S33: performing a commodity category texture feature analysis based on the commodity category target data to obtain commodity category texture feature data; Step S34: extracting surface subtle texture gradients based on the product category texture feature data to obtain product surface subtle texture gradient data; Step S35: Calculate the commodity category image feature vector according to the commodity surface subtle texture gradient data and the commodity category texture feature data to obtain the commodity category image feature vector data.

9. A commodity identification system based on an image classification model, characterized in that: Used to execute the commodity identification method based on the image classification model as claimed in claim 1, the commodity identification system based on the image classification model comprises: The physical parameter analysis module is used to obtain basic information data of commodities; perform physical parameter analysis of commodity categories according to the basic information data of commodities to obtain physical parameter data of commodity categories; The image processing module is used to collect commodity category images according to the commodity category physical parameter data, thereby obtaining commodity category image data; and perform commodity image contrast enhancement processing on the commodity category image data to obtain commodity category image contrast enhanced data; A feature vector calculation module is used to perform commodity target detection based on the commodity category image contrast enhancement data, thereby obtaining commodity category image target data; perform commodity category image feature vector calculation based on the commodity category image target data, thereby obtaining commodity category image feature vector data; The commodity classification and recognition module is used to analyze the subtle features of the external morphology of the commodity category based on the commodity category image feature vector data to obtain the subtle feature data of the external morphology of the commodity category; adjust the model recognition data of the subtle feature data of the external morphology of the commodity category based on the picture classification model to obtain the picture classification and recognition adjustment model; and perform commodity classification and recognition processing on the basic information data of the commodity based on the picture classification adjustment model to obtain commodity classification and recognition data.

10. A vending machine, characterized in that: It includes a vending machine main body, a power supply unit and an electrical control unit. The power supply unit is installed inside the vending machine main body. The electrical control unit is electrically connected to the power supply unit. The electrical control unit is used to charge and control the vending machine main body. The electrical control unit is used to execute the commodity recognition method based on the image classification model as described in any one of claims 1-8.

Images