Vegetable and fruit goods intelligent goods sorting and sorting system

Abstract

The application is suitable for the technical field of automatic detection and sorting of vegetables and fruits, and provides a kind of intelligent sorting system for vegetables and fruits, the system comprises: data acquisition module, for obtaining the collection image of the pitaya to be detected by polarized light imaging, and determining the initial abnormal index of the spot area in the collection image, while obtaining the surface properties of the specific area of the collection image;Image analysis module is used for analyzing the collection image obtained by polarized light imaging based on machine vision technology, and judging whether there is liquid drop phenomenon in the spot area, if there is, determine the corresponding liquid drop parameter.The application adopts deep learning and polarized light imaging technology, combines machine vision, and finely analyzes the spot area in the pitaya collection image, dynamically detects the influence of liquid drop on the reflection component of the spot, so as to realize the accurate correction of initial abnormal index.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of automatic detection and sorting of vegetables and fruits, and particularly relates to an intelligent vegetable and fruit goods sorting system. BACKGROUND

[0002] At present, the field of intelligent vegetable and fruit goods sorting mainly relies on traditional machine vision technology, which generates an abnormality index by collecting the overall reflectance characteristics of pitaya and uses it to determine the abnormality of the spot area. This method usually makes judgments based on fixed thresholds and overall image reflectance parameters, ignoring the differences in subtle local features in the pitaya spot area. Due to the natural color and texture variations on the surface of pitaya, combined with the influence of environmental factors such as light, angle and device parameters during the collection process, the overall reflectance parameters often cannot accurately reflect the true abnormality of the local spot area.

[0003] More importantly, the traditional method fails to fully consider the influence of tiny water droplets on the reflection components. After the droplets form on the surface of pitaya, they change the local reflectance characteristics, causing the reflection parameters to shift, and the fixed threshold cannot dynamically adapt to this change, resulting in a certain deviation in the calculation of the abnormality index. This defect is particularly prominent in practical applications, which may lead to the failure to accurately identify some abnormal spots, thereby affecting the overall detection accuracy of the sorting system and the product quality.

[0004] Therefore, the existing technology is difficult to achieve fine analysis of the abnormality of the pitaya spot area in complex environments and variable conditions, and there is an urgent need for a new method to dynamically capture local reflectance characteristics and accurately quantify the influence of factors such as droplets to make up for the shortcomings of the traditional fixed threshold method. SUMMARY

[0005] The purpose of the present application is to provide an intelligent vegetable and fruit goods sorting system to solve the problems raised in the background art.

[0006] The present application is implemented as follows: an intelligent vegetable and fruit goods sorting system, the system comprising:

[0007] a data acquisition module for acquiring a collection image of the pitaya to be inspected by polarized light imaging, and determining an initial abnormality index of the spot area in the collection image, while acquiring the surface properties of a specific area of the collection image;

[0008] an image analysis module for analyzing the collection image obtained by polarized light imaging based on machine vision technology, and determining whether there are droplets in the spot area, and if so, determining the corresponding droplet parameters;

[0009] a model application module, configured to call a predefined reference model, input surface properties of the specific area into the reference model for the first time and obtain an initial reference image, and then input the surface properties of the specific area and the droplet parameters into the reference model and obtain an additional reference image;

[0010] an anomaly index correction module, configured to determine a difference between the reflection components in the initial reference image and the additional reference image, and convert the difference into a correction coefficient to correct the initial anomaly index.

[0011] As a further limitation of the technical scheme of the embodiment of the present application, the image analysis module specifically comprises:

[0012] an image preprocessing unit, configured to pre-process the collected image obtained by imaging the polarized light by using machine vision technology, including image denoising, enhancement and edge detection, so as to highlight the detail features of the spot area and eliminate background interference;

[0013] a droplet phenomenon judgment unit, configured to analyze the spot area by using feature extraction and pattern recognition algorithms based on the pre-processed collected image, judge whether there is a droplet phenomenon by analyzing color, texture and reflected light characteristics, and confirm the existence of the droplet by using a set threshold value;

[0014] a droplet evaluation unit, configured to quantitatively evaluate the droplet in the spot area after confirming the existence of the droplet phenomenon, and extract the droplet density and thickness parameters.

[0015] As a further limitation of the technical scheme of the embodiment of the present application, the droplet evaluation unit specifically comprises:

[0016] a droplet contour extraction sub-unit, configured to accurately segment the droplet candidate area by using adaptive threshold segmentation or watershed algorithm combined with morphological operation, so as to realize the extraction of the droplet contour;

[0017] a droplet density determination sub-unit, configured to analyze the shape of the segmented droplet by using Canny edge detection and Hough circle transformation method, identify and measure the edge and diameter of the droplet, so as to determine the size of the droplet, and determine the droplet density by counting the ratio of the number of droplets in the spot area to the area of the spot area;

[0018] a droplet thickness calculation sub-unit, configured to detect the droplet by using a surface curvature analysis method based on three-dimensional reconstruction technology, so as to estimate the average thickness of the droplet.

[0019] As a further limitation of the technical scheme of the embodiment of the present application, the pre-defined reference model refers to an image generation and feature analysis model constructed based on a deep learning algorithm, which can generate a reference image based on the surface properties of the pitaya, and generate another reference image in combination with the surface properties of the pitaya and the droplet parameters, and generate corresponding reference reflection component parameters according to different surface properties of the pitaya, and generate modified reflection component parameters corresponding to the reference images generated after applying different droplet parameters.

[0020] As a further limitation of the technical scheme of the embodiment of the present application, the specific area of the collected image refers to an annular area outside the periphery of the spot area in the collected image, so as to ensure that the surface properties input to the pre-defined reference model are consistent with the surface properties of the spot area.

[0021] As a further limitation of the technical scheme of the embodiment of the present application, the abnormality index correction module specifically comprises:

[0022] The reflection component parameter determination unit is configured to determine the corresponding reference reflection component parameters and modified reflection component parameters in the initial reference image and the additional reference image based on the pre-defined reference model;

[0023] The correction coefficient acquisition unit is configured to quantify the difference between the reference reflection component parameters and the modified reflection component parameters, and convert the difference into a correction coefficient;

[0024] The abnormality index correction unit is configured to call a pre-set correction formula and correct the initial abnormality index according to the correction coefficient.

[0025] As a further limitation of the technical scheme of the embodiment of the present application, the pre-set correction formula is: , wherein refers to the corrected abnormality index, refers to the initial abnormality index, refers to the modified reflection component parameters, refers to the reference reflection component parameters, refers to the difference amount of the reflection components in the initial reference image and the additional reference image, i.e. the correction coefficient, refers to the adjustment coefficient of the correction coefficient.

[0026] Compared with the prior art, the present application has the following beneficial effects:

[0027] This invention employs deep learning and polarized light imaging technology, combined with machine vision, to perform detailed analysis of spotted areas in dragon fruit images. It dynamically detects the impact of droplets on the reflective components of the spots, thereby achieving precise correction of the initial anomaly index. This technology overcomes the limitations of traditional methods that rely solely on overall reflectance parameters and cannot accurately reflect spot anomalies. By comparing the differences in reflectance components of spotted areas between the initial and supplementary reference images, it effectively quantifies the droplet effect and generates correction coefficients. After implementing this invention, the sorting system can significantly improve the accuracy and robustness of detecting dragon fruit spot anomalies, ensuring product quality, reducing losses caused by detection errors, and providing a reliable and advanced technical solution for intelligent agricultural sorting. Attached Figure Description

[0028] Figure 1 Application architecture diagram of the system provided in the embodiments of the present invention;

[0029] Figure 2 This is a structural block diagram of the image parsing module in the system provided in the embodiments of the present invention;

[0030] Figure 3 This is a structural block diagram of the droplet evaluation unit in the system provided in the embodiments of the present invention;

[0031] Figure 4 This is a structural block diagram of the abnormal index correction module in the system provided in the embodiment of the present invention. Detailed Implementation

[0032] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0033] Furthermore, Figure 1 An application architecture diagram of the system provided in an embodiment of the present invention is shown.

[0034] In another preferred embodiment of the present invention, a smart sorting and distribution system for fruits and vegetables includes:

[0035] The data acquisition module 100 is used to acquire images of the dragon fruit to be inspected through polarized light imaging, determine the initial abnormality index of the spotted area in the acquired image, and acquire the surface properties of a specific area of ​​the acquired image.

[0036] In the embodiment of the present application, the data acquisition module 100 adopts a polarized light imaging technology, which captures the collected image of the fire dragon fruit to be inspected after illumination by a polarized light source through a high-resolution polarized camera equipped with a specific polarized filter, thereby simultaneously acquiring the intensity and polarization information of the image. This technical means relies on the cooperative work of the optical system, the polarized filter and the image acquisition software, and can effectively capture the information of the microstructure and moisture state of the fire dragon fruit surface.

[0037] The polarized light imaging method is mainly used because it can effectively suppress environmental light interference and glare, while enhancing the contrast of the target area in the image, providing more abundant surface details than traditional imaging, which is of great significance for distinguishing natural abnormalities in the fire dragon fruit spot area and abnormalities caused by liquid droplets.

[0038] The initial anomaly index in the collected image of the fire dragon fruit is obtained based on the existing mature machine vision and image processing technology by analyzing the color, texture and reflection characteristics of the spot area, and this method has been widely used in the fields of agricultural product quality detection, automatic sorting system and food processing.

[0039] The surface properties of the specific area of the collected image include color distribution, texture features, brightness distribution and reflection light characteristics, and also include polarization parameters (such as polarization angle and degree of polarization) obtained by polarized light imaging, which can reflect the microstructure and humidity information of the fire dragon fruit surface, providing data support for subsequent liquid droplet parameter extraction and anomaly index correction.

[0040] Further, the intelligent fruit and vegetable goods sorting system also comprises:

[0041] The image analysis module 200 is used to analyze the collected image obtained by polarized light imaging based on machine vision technology, and determine whether there is a liquid droplet phenomenon in the spot area, and if so, determine the corresponding liquid droplet parameters.

[0042] Specifically, Figure 2 The structure block diagram of the image analysis module 200 in the system provided by the embodiment of the present application is shown.

[0043] In the preferred embodiment provided by the present application, the image analysis module 200 specifically comprises:

[0044] The image preprocessing unit 201 is used to preprocess the collected image obtained by polarized light imaging by using machine vision technology, including image denoising, enhancement and edge detection, so as to highlight the detail features of the spot area and eliminate background interference;

[0045] The droplet phenomenon judging unit 202 is configured to judge whether the droplet phenomenon exists based on the preprocessed collected image, apply a feature extraction and pattern recognition algorithm to analyze the spot region, judge whether the droplet phenomenon exists by analyzing color, texture and reflection light characteristics, and confirm the existence of the droplet by using a set threshold value.

[0046] The droplet evaluation unit 203 is configured to quantitatively evaluate the droplet in the spot region after confirming the existence of the droplet phenomenon, and extract the droplet density and thickness parameters.

[0047] In the embodiment of the present application, the image preprocessing unit 201 first performs noise reduction processing on the collected image obtained by polarized light imaging, and commonly uses mature algorithms such as Gaussian filtering or median filtering to remove image noise; then the overall brightness and contrast of the image are improved by histogram equalization or adaptive contrast enhancement technology, and the detail features of the spot region are further highlighted; finally, the edge information in the image is extracted by using the Canny edge detection algorithm to effectively eliminate background interference and clearly define the spot region contour. These technologies are all mature machine vision means, and have been widely used in the field of industrial and agricultural detection.

[0048] After receiving the preprocessed image, the droplet phenomenon judging unit 202 first extracts the image features of the spot region by using color space conversion, local texture analysis (such as using local binary pattern or Gabor filtering) and reflection light characteristic analysis; then, according to the set threshold value or by using a trained deep learning model, the extracted features are recognized to judge whether the droplet phenomenon exists in the spot region; when the droplet phenomenon is detected, the corresponding droplet parameters are further confirmed and determined. The technical means based on this process also belong to the existing mature and achievable technology, and has been successfully applied in the field of automatic sorting and quality detection.

[0049] Specifically, Figure 3 The structural block diagram of the droplet evaluation unit 203 in the system provided by the embodiment of the present application is shown.

[0050] In the preferred embodiment provided by the present application, the droplet evaluation unit 203 specifically includes:

[0051] The droplet contour extraction subunit 2031 is configured to perform accurate segmentation on the droplet candidate region by using adaptive threshold segmentation or watershed algorithm combined with morphological operation, so as to realize the extraction of the droplet contour.

[0052] The droplet density determination subunit 2032 is configured to perform shape analysis on the segmented droplets by using a Canny edge detection and Hough circle transformation method, identify and measure the edges and diameters of the droplets to determine the sizes of the droplets, and determine the droplet density by calculating the ratio of the number of droplets in a spot area to the area of the spot.

[0053] The droplet thickness calculation subunit 2033 is configured to detect the droplets by using a surface curvature analysis method based on a three-dimensional reconstruction technology, and estimate the average thickness of the droplets.

[0054] In the embodiment of the present application, the droplet contour extraction subunit 2031 first uses an adaptive threshold segmentation method to preliminarily distinguish the candidate regions where the droplets may exist from the background in the collected image, and then further refines the segmentation result by using a watershed algorithm to solve the problem of droplet overlapping or adhesion. Then, the candidate regions are subjected to edge smoothing and noise suppression by combining morphological operations (such as dilation, erosion, opening operation and closing operation), so as to realize accurate extraction of the droplet contour. The above-mentioned techniques are all image segmentation methods widely used in the industry and academia, and can effectively cope with the challenges of complex textures and noise interference on the surface of pitaya.

[0055] The droplet density determination subunit 2032 uses a Canny edge detection algorithm to extract the edges of the segmented droplets in detail, and then uses a Hough circle transformation method to identify the circular features of the droplets and measure their diameters and edge information to determine the size of each droplet. By counting the number of droplets in a spot area and calculating the ratio of the number of droplets to the total area of the spot, the droplet density can be obtained. This process relies on mature edge detection and shape recognition algorithms to ensure that the distribution of droplets on the surface of pitaya can be accurately reflected under different image acquisition conditions.

[0056] The droplet thickness calculation subunit 2033 uses a surface curvature analysis method based on a three-dimensional reconstruction technology, combines multi-angle image acquisition or structured light technology, reconstructs the three-dimensional shape of the droplets, and analyzes the curvature change of the droplet surface to estimate the average thickness of the droplets. By introducing calibration parameters, this subunit can compensate for errors caused by viewing angle, illumination and droplet shape change, and improve the accuracy of thickness measurement. The overall technical solution is based on existing mature three-dimensional imaging and analysis means, and provides reliable support for quantitative evaluation of the thickness of pitaya droplets.

[0057] Further, the vegetable and fruit product intelligent sorting system further comprises:

[0058] The model application module 300 is configured to call a predefined reference model, input the surface properties of a specific region into the reference model for the first time and obtain an initial reference image, and simultaneously input the surface properties of the specific region and the droplet parameters into the reference model and obtain an additional reference image.

[0059] The predefined reference model refers to an image generation and feature analysis model constructed based on a deep learning algorithm, which can generate a reference image based on the surface properties of pitaya, generate another reference image in combination of the surface properties of pitaya and droplet parameters, generate corresponding reference reflection component parameters according to different surface properties of pitaya, and generate modified reflection component parameters corresponding to the reference image generated after applying different droplet parameters.

[0060] The specific region of the collected image refers to the annular region outside the periphery of the spot region in the collected image, so as to ensure that the surface properties input into the predefined reference model are consistent with the surface properties of the spot region.

[0061] In the embodiment of the present application, the predefined reference model is constructed based on a deep learning algorithm, and the model is trained by using a large amount of pitaya image data, so as to realize the functions of image generation and feature analysis. The model can adopt mature technologies such as convolutional neural network (CNN) or generative adversarial network (GAN), and is constructed based on existing deep learning frameworks such as TensorFlow and PyTorch. By collecting and labeling image data of pitaya under different surface property conditions, the model can learn the corresponding reference reflection component parameters of different surface properties; when the droplet parameters are input, the model generates a modified reference image reflecting the influence of droplets, and outputs corresponding modified reflection component parameters.

[0062] The model application module 300 first calls the predefined reference model, and inputs the surface properties reflected by the annular region outside the periphery of the spot region in the collected image into the model to generate an initial reference image, which reflects the reference reflection component parameters of pitaya under the influence of droplets. Then, the same surface properties and droplet parameters are input into the model to generate an additional reference image, which reflects the influence of the presence of droplets on the reflection component. By comparing the reflection component parameters in the initial reference image and the additional reference image, the correction effect of droplets on the reflection characteristics of the surface of pitaya can be quantitatively calculated.

[0063] Further, the vegetable and fruit commodity intelligent sorting system further comprises:

[0064] The anomaly index correction module 400 is configured to determine the difference between the reflection components in the initial reference image and the additional reference image, and convert the difference into a correction coefficient to correct the initial anomaly index.

[0065] Specifically, Figure 4 The structure block diagram of the anomaly index correction module 400 in the system provided by the embodiment of the present application is shown.

[0066] In the preferred embodiment provided by the present application, the anomaly index correction module 400 specifically comprises:

[0067] The reflection component parameter determination unit 401 is configured to determine the reference reflection component parameter and the corrected reflection component parameter in the initial reference image and the additional reference image respectively based on a predefined reference model.

[0068] The correction coefficient acquisition unit 402 is configured to quantify the difference between the reference reflection component parameter and the corrected reflection component parameter and convert the difference into a correction coefficient.

[0069] The abnormality index correction unit 403 is configured to call a preset correction formula and correct the initial abnormality index according to the correction coefficient.

[0070] The preset correction formula is as follows: wherein denotes the corrected abnormality index, denotes the initial abnormality index, denotes the corrected reflection component parameter, denotes the reference reflection component parameter, denotes the difference between the reflection components in the initial reference image and the additional reference image, i.e., the correction coefficient, denotes the adjustment coefficient of the correction coefficient.

[0071] In the prior art, the generation of the initial abnormality index generally only depends on the overall reflection characteristics of the pitaya spot area, without fully considering the influence of the tiny water droplets on the reflection component or accurately quantifying the influence, resulting in a possible deviation between the abnormality index and the actual abnormal condition.

[0072] To solve this problem, the abnormality index correction module 400 of the present application adopts a predefined reference model to extract the corresponding reflection component parameters from the initial reference image reflecting only the surface properties of the pitaya and the additional reference image reflecting both the surface properties and the droplet parameters.

[0073] By comparing the difference between the two, i.e., quantifying the change in the reflection component caused by the presence of droplets, the difference is converted into a correction coefficient to dynamically correct the initial abnormality index, thereby more accurately reflecting the actual abnormal condition. This method takes advantage of the deep learning model to generate reference images, making up for the deficiency in the prior art that cannot accurately capture and quantify the influence of tiny water droplets on the reflection characteristics, significantly improving the scientificity of the abnormality index and the robustness of the system.

[0074] The basic idea of the preset correction formula is to first calculate the relative difference between the reference reflection component parameters extracted from the initial reference image and the correction reflection component parameters extracted from the additional reference image, which reflects the influence of droplets and other factors on the reflection characteristics of pitaya. Then, multiply this relative difference (correction coefficient) by a preset adjustment coefficient to obtain a correction factor, which is used to dynamically correct the initial anomaly index, so that the corrected anomaly index can more accurately reflect the actual abnormal condition. Such setting is because the existing technology usually fails to fully consider the influence of small droplets on the reflection component when generating the initial anomaly index, while the present method can intuitively reflect the deviation caused by droplets by quantitatively comparing the reflection parameters under droplet-free and droplet conditions, thereby reasonably correcting the anomaly index.

[0075] It should be noted that the correction formula is only a simple and intuitive calculation method, and in actual application, more advanced and complex calculation methods can be used according to requirements. For example, a nonlinear regression model or a neural network-based dynamic adjustment algorithm can be used to model and predict the influence of droplets more accurately through machine learning, or an adaptive filtering method can be used to update the adjustment coefficient in real time, to further improve the accuracy and robustness of anomaly index correction.

[0076] It should be understood that although each step in the flowchart of each embodiment of the present application is displayed in sequence according to the arrow, these steps are not necessarily executed in the order indicated by the arrow. Unless otherwise specified herein, the execution of these steps is not strictly limited in sequence, and these steps can be executed in other orders. Moreover, at least part of the steps in each embodiment can include multiple sub-steps or multiple stages, which are not necessarily executed at the same time, but can be executed at different times, and the execution order of these sub-steps or stages is not necessarily sequential, but can be alternately executed with at least part of other steps or sub-steps or stages of other steps.

[0077] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program. The program can be stored in a non-volatile computer readable storage medium, and when the program is executed, the processes of the above-mentioned embodiments of the methods can be included. Any reference to memory, storage, database, or other medium used in the embodiments provided in the present application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration but not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0078] Any combination of the technical features of the above-mentioned embodiments can be combined. In order to make the description simple, all possible combinations of the technical features in the above-mentioned embodiments are not described, however, as long as the combination of the technical features does not exist contradictory, it should be considered as the scope of the present application.

[0079] The above-mentioned embodiments only express several embodiments of the present application, and the description is more specific and detailed, but it should not be understood as limiting the scope of the present application. It should be noted that for ordinary skilled in the art, without departing from the concept of the present application, a number of modifications and improvements can be made, which are within the scope of the present application. Therefore, the protection scope of the present application should be subject to the appended claims.

[0080] The above-mentioned is only the preferred embodiment of the present application, and does not limit the present application. Any modification, equivalent replacement, and improvement made within the spirit and principle of the present application should be included in the protection scope of the present application.

Claims

1. A smart sorting and distribution system for fruits and vegetables, characterized in that, The system includes: The data acquisition module is used to acquire images of the dragon fruit to be inspected through polarized light imaging, determine the initial abnormality index of the spotted area in the acquired image, and acquire the surface properties of a specific area of ​​the acquired image. The image analysis module is used to analyze the acquired image obtained by polarized light imaging based on machine vision technology, and to determine whether there is a droplet phenomenon in the spot area. If so, the corresponding droplet parameters are determined. The model application module is used to retrieve a predefined reference model. Initially, the surface properties of a specific region are input into it to obtain an initial reference image. Then, the surface properties of the specific region and the droplet parameters are input into it simultaneously to obtain an additional reference image. The anomaly index correction module is used to determine the difference in reflectance components between the initial reference image and the additional reference image, and convert it into correction coefficients to correct the initial anomaly index. The abnormal index correction module specifically includes: The reflection component parameter determination unit is used to determine the reference reflection component parameters and the modified reflection component parameters in the initial reference image and the additional reference image, respectively, based on a predefined reference model. The correction coefficient acquisition unit is used to quantify the difference between the reference reflection component parameters and the corrected reflection component parameters, and convert the difference into correction coefficients. Anomaly index correction unit is used to retrieve a preset correction formula and correct the initial anomaly index according to the correction coefficient; The preset correction formula is: ,in This refers to the corrected anomaly index. This refers to the initial anomaly index. This refers to correcting the reflection component parameters. This refers to the reference reflectance component parameters. This refers to the difference in reflectance between the initial reference image and the additional reference image, i.e., the correction factor. This refers to the adjustment coefficient of the correction coefficient.

2. The intelligent sorting and distribution system for fruits and vegetables according to claim 1, characterized in that, The image parsing module specifically includes: The image preprocessing unit is used to preprocess the acquired images obtained by polarized light imaging using machine vision technology, including image denoising, enhancement and edge detection, to highlight the detailed features of the spot area and eliminate background interference. The droplet phenomenon judgment unit is used to analyze the spot area based on the preprocessed acquired image by applying feature extraction and pattern recognition algorithms. It determines whether there is a droplet phenomenon by analyzing the color, texture and reflected light characteristics, and confirms the existence of droplets by using a set threshold. The droplet assessment unit is used to quantitatively assess the droplets in the spot area after confirming the presence of droplets, and to extract droplet density and thickness parameters.

3. The intelligent sorting and distribution system for fruits and vegetables according to claim 2, characterized in that, The droplet evaluation unit specifically includes: The droplet contour extraction subunit is used to accurately segment the candidate droplet regions in the preprocessed acquired image by using adaptive threshold segmentation or watershed algorithm combined with morphological operations to extract the droplet contour. The droplet density determination subunit is used to perform shape analysis on the segmented droplets using Canny edge detection and Hough circle transform methods, identify and measure the edges and diameters of the droplets to determine the size of the droplets, and determine the droplet density by statistically analyzing the ratio of the number of droplets in the spot region to the area of ​​the region. The droplet thickness calculation subunit is used to detect droplets using a surface curvature analysis method based on three-dimensional reconstruction technology in order to estimate the average thickness of the droplets.

4. The intelligent sorting and distribution system for fruits and vegetables according to claim 1, characterized in that, The predefined reference model refers to an image generation and feature analysis model built based on a deep learning algorithm. It can generate a reference image based on the surface properties of dragon fruit, and generate another reference image by combining the surface properties of dragon fruit and droplet parameters. At the same time, it generates corresponding baseline reflectance component parameters according to different dragon fruit surface properties, and generates corrected reflectance component parameters corresponding to the reference images generated after applying different droplet parameters.

5. The intelligent sorting and distribution system for fruits and vegetables according to claim 4, characterized in that, The specific region of the acquired image refers to the annular region surrounding the spot region in the acquired image, to ensure that the surface properties subsequently input to the predefined reference model are consistent with the surface properties of the spot region.

Images