Non-destructive testing method, device and system for internal defects of fruit

Abstract

The application discloses a nondestructive detection method, device and system for fruit water core disease, and solves the problems of the existing water core disease detection technology in cost, efficiency, robustness, sorting accuracy and industrial application. In step S1, multiple images of the fruit to be detected are obtained, and image preprocessing and correction are performed on each image one by one; in step S2, the color feature, area feature and texture feature of the fruit to be detected are extracted in the effective peel area of each image; in step S3, the fruit variety of the fruit to be detected is obtained, the pre-stored weight coefficient is called, and the composite yellowing index FI is constructed based on the color feature, area feature, texture feature of the fruit to be detected and the weight coefficient; in step S4, the yellowing index FI of the fruit to be detected in each image is calculated, and compared with the preset threshold one by one, so that each image of the fruit to be detected is classified into a healthy fruit, a suspicious fruit or a water core disease fruit.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of fruit detection, in particular to a non-destructive detection method, device and system for water core disease of fruit. BACKGROUND

[0002] Water core disease is a common physiological disease of fruits such as pears and apples. After the disease occurs, the fruit flesh appears translucent vitrification, accompanied by yellowing of the fruit skin, which seriously damages the fruit quality and commodity value. Therefore, it is crucial to achieve rapid and accurate identification of water core disease in the post-harvest sorting process of fruits.

[0003] Currently, the detection methods for water core disease mainly include the following two categories: destructive detection and physical field non-destructive detection.

[0004] 1. Destructive detection method: mainly represented by manual cutting and observation. Although this method is intuitive, it is destructive and cannot be applied to commercial sorting processes, and is inefficient.

[0005] 2. Physical field non-destructive detection technology: mainly includes near-infrared transmission spectroscopy, X-ray or CT imaging, and ultrasonic detection. Although this type of method can achieve non-destructive detection, it has the following obvious limitations:

[0006] High equipment cost: near-infrared spectrometer, X-ray CT, etc. are expensive and have high maintenance costs.

[0007] Detection efficiency is limited: the detection speed is usually slow, which is difficult to meet the high-throughput requirements of modern sorting lines.

[0008] There are application bottlenecks: X-ray technology has radiation safety hazards; ultrasonic detection requires coupling agents and is easily affected by fruit shape and position, with poor robustness.

[0009] Among these technologies, for example, patent document CN109100323A discloses a non-destructive quantitative evaluation method for apple water core disease using transmission spectroscopy. This method collects the near-infrared transmission spectrum of the apple, and needs to cut the apple to accurately calculate the water core disease area ratio using image segmentation algorithm as the modeling benchmark, and then establishes a quantitative relationship model between the spectrum and the internal disease degree. Although this method can reflect the internal information of the fruit, it has its own limitations that are difficult to overcome: (1) the equipment is complex and expensive, it depends on precise instruments, which is not conducive to large-scale sorting line promotion; (2) the instrument needs to be calibrated frequently and has high environmental requirements, which is not suitable for continuous detection of production lines; (3) the modeling process is complex, and a multivariate correction model must be established through a large amount of "spectrum data + cross-section image", which leads to poor model migration between different batches and different varieties; (4) the detection process is long, the spectrum needs to be measured at multiple points, and then complex preprocessing and modeling prediction are required, which takes a long time for single fruit detection, making it difficult to adapt to high-speed sorting lines.

[0010] 3. Machine vision-based detection methods: In recent years, some studies have attempted to use RGB cameras to acquire images of fruit surfaces and identify yellowing areas associated with watercore disease by setting color thresholds. This method is low-cost, but it has the following fundamental drawbacks.

[0011] It has limited features and poor robustness: relying solely on epidermal color information, it is highly susceptible to interference from irrelevant factors such as changes in ambient light, sunspots on the fruit surface, and mechanical damage, leading to a high misjudgment rate.

[0012] Limited identification accuracy: Most studies use a simple dichotomy of "healthy" and "disease-prone", which cannot effectively distinguish between "healthy fruit", "suspicious fruit" and "disease-prone fruit", and lacks the ability to handle fruit in intermediate state, which limits the improvement of sorting accuracy.

[0013] Insufficient industrial application: Existing research is mostly still in the laboratory stage, and there is a lack of mature equipment that can adapt to assembly line operation environments and integrate online calibration and automatic sorting functions.

[0014] In summary, existing watercore detection technologies have shortcomings in terms of cost, efficiency, robustness, sorting accuracy, and industrial application. Therefore, there is an urgent need for a non-destructive testing method and equipment that can balance low cost, high efficiency, and high accuracy, and is suitable for industrialized production lines, to address the practical technical challenges in detecting watercore in fruits such as green-skinned pears. Summary of the Invention

[0015] This application provides a non-destructive testing method, apparatus, and system for watercore disease in fruits, aiming to solve the shortcomings of existing watercore disease detection technologies in terms of cost, efficiency, robustness, sorting accuracy, and industrial application.

[0016] Firstly, a non-destructive detection method for watercore disease in fruit is provided, comprising the following steps S1-S5.

[0017] Step S1: Acquire multiple images of the fruit to be detected, and perform image preprocessing and correction on each image; wherein, the multiple images record multiple angles of the fruit to be detected.

[0018] Step S2: Extract the color features, area features, and texture features of the fruit to be detected within the effective skin area of ​​each image.

[0019] Step S3: Obtain the fruit variety of the fruit to be tested, call the pre-stored weight coefficients and threshold parameters for grading and discrimination, and construct a composite yellowing index FI based on the color features, area features, texture features of the fruit to be tested and the weight coefficients, so as to comprehensively reflect the yellowing degree and tissue uniformity of the fruit to be tested.

[0020] In step S4, the yellowing index FI of the fruit to be detected in each image is calculated by using the constructed composite yellowing index FI, and is compared with a preset threshold one by one, so as to classify each image of the fruit to be detected as a healthy fruit, a suspicious fruit or a fruit with internal browning disease.

[0021] In step S5, the calculation results of each image of the fruit to be detected are fused, and a preliminary classification result of the fruit to be detected is output, and the preliminary classification result is specifically a healthy fruit, a suspicious fruit or a fruit with internal browning disease.

[0022] In the above method, optionally, in step S3, the composite yellowing index FI is specifically the following formula.

[0023] .

[0024] In the above formula, , , represent the weight coefficients calibrated by training samples; , , represent the color features of the fruit to be detected extracted, where , represent the CIELab components, represents the HSV hue component; represents the area feature of the fruit to be detected extracted, that is, the proportion of the yellowing area; represents the texture feature of the fruit to be detected extracted, that is, the texture uniformity index.

[0025] In the above method, optionally, step S4 specifically includes: comparing the calculated composite yellowing index FI with a preset threshold to achieve three classifications of the fruit to be detected: healthy fruit: FI < T1; suspicious fruit: T1 ≤ FI ≤ T2; fruit with internal browning disease: FI > T2; where the first preset threshold T1 and the second preset threshold T2 are obtained by training with experimental data and are adaptively corrected according to batch differences, and the dynamic correction rule of the threshold is the following formula.

[0026] .

[0027] In the above formula, represents the mean value of healthy fruits in the current batch, represents the reference mean value, represents the correction coefficient; represents the dynamically corrected threshold; represents the base threshold.

[0028] Optionally, step S1 in the above method includes: acquiring multi-angle images of the fruit to be tested using an RGB industrial camera; performing white balance processing and color correction on each of the acquired images; performing reference calibration using a standard gray card or color card; and converting the images to the CIELab and HSV color spaces.

[0029] Optionally, in the above method, step S5 specifically includes: fusing the calculation results of each image of the fruit to be detected using a voting or probability weighting method, and obtaining the preliminary classification result of the fruit to be detected.

[0030] Optionally, after step S5, the method further includes: performing single-band near-infrared detection or manual verification on the fruit to be tested for which the preliminary classification result is suspicious, so as to obtain the final classification result.

[0031] Secondly, a non-destructive testing device for watercore disease in fruit is provided. The device is used to perform the above-mentioned testing method and includes: an optical detection module, an image acquisition mechanism, a calibration module, a data processing unit, and a sorting mechanism.

[0032] An optical inspection module includes: a closed light box with diffuse reflection on its inner wall; a ring-shaped LED light source arranged on the top of the inner wall of the closed light box, which is a multi-angle illumination array; an RGB industrial camera arranged inside the closed light box for photographing the fruit to be inspected; and a cross polarizer arranged between the ring-shaped LED light source and the RGB industrial camera to suppress surface highlights.

[0033] The image acquisition mechanism, specifically a turntable or multi-camera arrangement structure, is used to feed the fruit to be tested into the optical detection module, and cooperate with the optical detection module to realize multi-angle image acquisition of the fruit to be tested.

[0034] The calibration module is used to calibrate the multi-angle images of the fruit to be tested. It has a built-in gray card or color card and is connected to the RGB industrial camera. It automatically corrects the image when the device is turned on, realizing online white balance and color calibration.

[0035] The data processing unit, including an embedded processor or edge computing platform, is used to perform image preprocessing, composite yellowing index (FI) calculation, fruit classification algorithm, and threshold correction logic. Simultaneously, it supports communication with the cloud to receive updated parameters sent from the cloud, including weighting coefficients for calculating the composite yellowing index and / or threshold parameters for grading.

[0036] The sorting mechanism is a three-channel diversion device, including: a healthy fruit screening channel, a suspicious fruit screening channel, and a water-core disease removal channel.

[0037] Optionally, in the above method, the suspicious fruit screening channel of the sorting mechanism is connected to a secondary detection module or a manual verification port.

[0038] Thirdly, a non-destructive testing system for watercore disease in fruit is provided, comprising: a first-level detection unit and a second-level detection unit.

[0039] The first-level detection unit is used to perform the above-mentioned detection methods to conduct preliminary detection and classification of fruits.

[0040] The second-level detection unit performs single-band near-infrared detection or manual verification on the suspicious fruits obtained from the first-level detection unit; wherein, the system performs final classification of the suspicious fruits based on the verification results of the second-level detection unit.

[0041] Optionally, the above method may also include: edge computing nodes and cloud service platforms.

[0042] An edge computing node is deployed at the detection site, and the first-level detection unit is integrated on the edge computing node to realize the real-time calculation of the detection method.

[0043] A cloud service platform, which communicates with the edge computing node, is used to receive and store detection data, perform batch difference analysis and update parameters, and send the updated parameters to the edge computing node; wherein, the updated parameters include weighting coefficients for calculating the composite yellowing index and / or threshold parameters for grading discrimination.

[0044] Compared with the prior art, this application has at least the following beneficial effects.

[0045] This application, based on further analysis and research into the problems of existing technologies, recognizes that current watercore detection technologies have shortcomings in terms of cost, efficiency, robustness, sorting accuracy, and industrial application. By replacing expensive dedicated instruments with conventional imaging equipment, the application reduces equipment investment costs. Simultaneously, it acquires multi-angle images of the fruit to be tested and performs preprocessing and correction to eliminate environmental interference and improve robustness. By simultaneously extracting multi-dimensional features of color, area, and texture within the effective peel area, it avoids the limitations of single features. By combining the variety of the fruit to be tested with pre-stored weighting coefficients, it calculates a composite yellowing index (FI) that comprehensively reflects the degree of yellowing and tissue uniformity, adapting to varietal differences and improving the comprehensiveness of feature representation. Through single-image classification and multi-image result fusion, it further improves discrimination accuracy. The entire process is adaptable to automated processing and can be directly integrated into existing industrial sorting production lines. This significantly reduces detection costs, improves detection efficiency and environmental adaptability, enhances the sorting accuracy of watercore-affected fruit, and achieves stable adaptation to industrial production scenarios, effectively solving multiple pain points of existing technologies in practical applications.

[0046] The solution proposed in this application also has at least the following effects.

[0047] I. Innovation and advantages at the methodological level.

[0048] Low detection cost and easy to promote: The core detection method of this application only requires an RGB camera and an LED light source, which completely avoids the dependence on high-cost and large-scale equipment such as near-infrared spectroscopy, X-ray, and CT, greatly reducing the technical threshold and system cost, and laying the foundation for large-scale promotion in fruit production areas, sorting plants and other links.

[0049] Strong robustness and high accuracy in identification: The composite yellowing index proposed in this application innovatively integrates CIELab / HSV color features, yellowing area ratio and texture uniformity index, which can effectively distinguish yellowing caused by watercore disease from epidermal abnormalities such as sun spots and mechanical damage, significantly improving the accuracy of discrimination and anti-interference ability.

[0050] The grading strategy is sophisticated and highly practical: it adopts a three-tier grading mechanism of healthy fruit, suspicious fruit, and water-core diseased fruit, which breaks through the traditional simple dichotomy of "qualified / unqualified". It not only greatly improves the sorting accuracy, but also provides a secondary verification channel for suspicious fruit, effectively reducing the misjudgment rate and has high practical value in the industry.

[0051] Strong adaptability and excellent stability: The dynamic threshold correction rule introduced in this application can adaptively adjust the discrimination criteria according to the average characteristics of different batches of fruit, effectively addressing the appearance changes caused by differences in orchard origin, harvesting time, and storage conditions, and ensuring the stability of the detection method in long-term application.

[0052] II. Innovation and advantages at the device level.

[0053] Specialized structural design and high degree of industrialization: The dedicated detection device of this application integrates a cross-polarized light source (effectively suppressing epidermal highlights), a multi-angle imaging mechanism (achieving blind-spot-free detection of the entire fruit surface), and a built-in calibration module (ensuring long-term color consistency), ensuring the reliability of detection results from a hardware perspective. Combined with a three-channel sorting mechanism, it can achieve seamless integration with production lines, automatically completing fruit sorting with high detection and sorting efficiency, possessing direct industrial application value.

[0054] III. Innovation and advantages at the system level.

[0055] The architecture is optimized to balance efficiency and accuracy: the two-level detection system constructed in this application adopts a collaborative strategy of "RGB main detection + near-infrared / manual verification". The first level uses RGB detection to achieve rapid screening at low cost and high efficiency; the second level performs precise verification for suspicious results. This architecture achieves the best balance between detection accuracy and processing efficiency on industrial production lines.

[0056] The system is intelligently expandable and capable of evolution: it supports edge computing and cloud-based collaboration, allowing detection data to be uploaded to the cloud for batch analysis and model optimization. Updated parameters can also be periodically distributed to the edge, enabling the system to learn and continuously optimize. Furthermore, the core principles of this solution mean it is not only applicable to green-skinned pears but can also be extended to the detection of internal diseases in other pear varieties, as well as apples, citrus fruits, and other fruits, demonstrating broad versatility and promising prospects for wider adoption. Attached Figure Description

[0057] Figure 1 This is a flowchart illustrating a non-destructive testing method for watercore disease in fruit, provided in the first embodiment of this application.

[0058] Figure 2 This is a schematic diagram of the algorithm logic of a non-destructive testing method for watercore disease in fruit provided in one embodiment of this application.

[0059] Figure 3 This is a flowchart illustrating a non-destructive testing method for watercore disease in fruit, provided as a second embodiment of this application.

[0060] Figure 4 This is a schematic diagram of a non-destructive testing device for watercore disease in fruit, provided as an embodiment of this application.

[0061] Figure 5 This is a schematic diagram of an RGB detection dark box for a non-destructive testing device for watercore disease in fruit, provided as an embodiment of this application.

[0062] Figure 6 This is a structural diagram of the internal structure of the RGB detection dark box of a non-destructive testing device for watercore disease in fruit provided in one embodiment of this application.

[0063] Figure 7 This is a partial detail of the near-infrared re-examination dark box of a non-destructive testing device for watercore disease in fruit provided in one embodiment of this application.

[0064] Among them, 1, electric push rod; 2, sorting trough; 3, fruit moving platform; 4, RGB detection dark box; 4-1, terminal controller; 4-2, light-blocking curtain; 4-3, ring light source; 4-4, circular track; 4-5, multi-degree-of-freedom camera joystick; 4-6, industrial camera; 4-7, calibration color card; 5, near-infrared re-inspection dark box; 5-1, near-infrared camera. Detailed Implementation

[0065] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0066] In the description of this application: unless otherwise stated, "a plurality of" means two or more. The terms "first," "second," "third," etc., in this application are intended to distinguish the objects referred to and do not have any special meaning in terms of technical connotation (e.g., they should not be construed as an emphasis on importance or order). Expressions such as "including," "comprising," and "having" also mean "not limited to" (certain units, components, materials, steps, etc.).

[0067] Green-skinned pears (such as Su Cui No. 1) come in many varieties and are of high quality, but they are highly susceptible to watercore disease, which causes the flesh to vitrify and the peel to turn yellow, severely impacting their commercial value. Existing detection methods mainly include destructive methods involving cutting the fruit, near-infrared spectroscopy, X-ray or CT imaging, and ultrasonic testing. While cutting is intuitive, it damages the fruit and cannot be used for commercial sorting. Near-infrared and X-ray methods offer some accuracy, but the equipment is expensive, bulky, and slow, making it difficult to meet the needs of large-scale applications in fruit production lines, and also poses safety risks. Ultrasonic testing requires contact operation, is inefficient, and is easily misjudged due to the shape and position of the fruit. In recent years, there have been attempts to identify fruit yellowing using RGB image color thresholds, but this method relies solely on changes in peel color, making it susceptible to interference from ambient light, angle, and sunspots on the peel, resulting in insufficient robustness. Moreover, most research remains at the laboratory stage, lacking mature industrial equipment support. Furthermore, existing solutions often employ simple binary discrimination, failing to achieve precise grading of healthy, suspected, and diseased fruit, limiting sorting accuracy and subsequent verification processes. It is evident that existing technologies are inadequate in terms of cost, efficiency, robustness, and equipment application. There is an urgent need for a low-cost, high-accuracy, and production-line-suitable non-destructive testing method and supporting equipment to address the technical challenges in detecting water core disease in green pears.

[0068] Therefore, the purpose of this scheme is to propose a composite yellowing index method based on the fusion of RGB color and texture features, combined with a three-level classification and an adaptive threshold correction mechanism, which can detect water core disease in green-skinned pears in a low-cost, high-efficiency and robust manner.

[0069] In one embodiment, a non-destructive detection method for watercore disease in fruit is provided, comprising the following steps S1-S5.

[0070] Step S1: Acquire multiple images of the fruit to be detected, and perform image preprocessing and correction on each image; wherein, the multiple images record multiple angles of the fruit to be detected.

[0071] In one embodiment, step S1 specifically includes: using an RGB industrial camera to acquire multi-angle images of the fruit to be inspected.

[0072] Each of the acquired images was subjected to white balance processing and color correction. The images were then calibrated using a standard gray card or color card and converted to the CIELab and HSV color spaces.

[0073] In this embodiment, multi-angle images of the fruit are captured using an RGB camera under a constant color temperature light source, and high light reflection is suppressed by a cross-polarization optical structure; the images are then white-balanced and color-corrected, and converted to CIELab and HSV color spaces.

[0074] Step S2: Extract the color features, area features, and texture features of the fruit to be detected within the effective skin area of ​​each image.

[0075] Step S3: Obtain the fruit variety of the fruit to be tested, call the pre-stored weight coefficients and threshold parameters for grading and discrimination, and construct a composite yellowing index FI based on the color features, area features, texture features of the fruit to be tested and the weight coefficients, so as to comprehensively reflect the yellowing degree and tissue uniformity of the fruit to be tested.

[0076] In one embodiment, in step S3, the composite yellowing index FI is specifically defined as follows.

[0077] .

[0078] In the above formula, , , This represents the weight coefficients calibrated using the training samples; , , This represents the extracted color features of the fruit to be tested, where, , Represents the CIELab component. Indicates HSV hue components; This indicates the area characteristics of the extracted fruit to be tested, namely the proportion of yellowed area; This represents the extracted texture features of the fruit to be tested, i.e., the texture uniformity index.

[0079] Step S4: Using the constructed composite yellowing index FI, calculate the yellowing index FI of the fruit to be detected in each image, and compare it with a preset threshold one by one, thereby classifying each image of the fruit to be detected into healthy fruit, suspicious fruit, or water core diseased fruit.

[0080] In one embodiment, step S4 specifically includes: comparing the calculated composite yellowing index FI with a preset threshold to achieve three classifications of the fruits to be detected: (1) healthy fruits: FI < T1; (2) suspicious fruits: T1 ≤ FI ≤ T2; (3) fruits with core rot: FI > T2; where the first preset threshold T1 and the second preset threshold T2 are obtained through training with experimental data and are adaptively corrected according to batch differences. The dynamic correction rule of the threshold is as follows.

[0081] .

[0082] In the above formula, represents the mean value of healthy fruits in the current batch, represents the reference mean value, represents the correction coefficient; represents the dynamically corrected threshold; represents the base threshold.

[0083] In this embodiment, (Base threshold): This is the standard threshold determined through statistics and modeling in a large number of sample experiments in the early stage. It is the classification discrimination boundary (such as T1, T2) obtained under benchmark conditions (such as standard light, fruits of a standard batch, mean ). It reflects the ideal threshold for core rot detection under "reference conditions".

[0084] (Dynamically corrected threshold): In actual sorting applications, due to differences in maturity, light environment, and orchard source among fruits of different batches, the overall distribution of the peel color (mean ) will shift. If is directly used, discrimination deviation may occur. Therefore, it is necessary to correct the base threshold according to the mean shift amount of this batch (

[0087] , ) to obtain a new threshold applicable to the current batch.

[0085] Step S5: Fuse the calculation results of each image of the fruits to be detected, and output the preliminary classification result of the fruits to be detected. The preliminary classification result is specifically a healthy fruit, a suspicious fruit, or a fruit with core rot.

[0086] In one embodiment, step S5 specifically includes: fusing the calculation results of each image of the fruits to be detected by means of voting or probability weighting, and the preliminary classification result of the fruits to be detected.

[0087] In one embodiment, after step S5, the method further includes: performing single-band near-infrared detection or manual verification on the fruit to be tested for which the initial classification result is suspicious, thereby obtaining the final classification result.

[0088] This application discloses a non-destructive detection method and device for watercore disease in green-skinned pears (such as Su Cui No. 1). The method is based on RGB image acquisition and color space analysis, utilizing a cross-polarized light source and multi-angle imaging to acquire fruit epidermal information. After white balance and color correction, the image is converted to CIELab and HSV spaces to extract color, area, and texture features, constructing a composite yellowing index. Based on dynamic threshold correction and a three-tier grading mechanism, the fruit is classified into three categories: healthy, suspected, and watercore. The device consists of an optical detection module, a circular track multi-angle acquisition mechanism, a built-in calibration module, a data processing unit, and a three-channel sorting mechanism, supporting a two-level detection architecture of RGB detection and near-infrared verification. This application has advantages such as low cost, high accuracy, and online application, making it suitable for fruit sorting and commercial processing.

[0089] The following describes the scheme of this application from another perspective.

[0090] 1. Image acquisition.

[0091] The fruit to be inspected is placed in a closed optical inspection module, illuminated by a ring-shaped LED light source with a constant color temperature (5000–5500K). Cross polarizers are placed at the light source and camera ends to suppress high-gloss reflections from the fruit skin. An RGB industrial camera is used to acquire images of the fruit from multiple angles, achieving coverage in at least three directions through a circular track arrangement.

[0092] 2. Preprocessing and correction.

[0093] The acquired images undergo white balance processing and color correction, are calibrated using a standard gray card or color card, and are converted to the CIELab and HSV color spaces.

[0094] 3. Target segmentation.

[0095] Extract color features within the effective skin area ( , , Area characteristics (percentage of yellowed area) ) and texture features (texture uniformity) ).

[0096] 4. Feature extraction.

[0097] Input the types of fruits to be stored and the required parameters.

[0098] 5. Calculation of the compound yellowing index.

[0099] Construct a composite yellowing index (FI) according to the following formula to comprehensively reflect the degree of fruit yellowing and tissue uniformity.

[0100] .

[0101] In the above formula, , , are weight coefficients calibrated through training samples.

[0102] 6. Grading discrimination.

[0103] Compare the calculated FI with a set threshold to achieve three-grade classification: healthy fruit: FI < T1; suspicious fruit: T1 ≤ FI ≤ T2; water core fruit: FI > T2.

[0104] The thresholds T1 and T2 can be obtained through training with experimental data and are adaptively corrected according to batch differences: ; where is the reference mean of the current batch, is the standard mean, is the correction coefficient.

[0105] 7. Result fusion and output.

[0106] Fuse the results from multiple angles using voting or probability weighting, and output the final discrimination grade. The discrimination result is linked to the sorting mechanism to automatically divert the fruits to the corresponding channels.

[0107] In one embodiment, a non-destructive detection device for fruit water core disease is provided. The device is used to complete the detection method provided in the above embodiment and includes: an optical detection module, an image acquisition mechanism, a calibration module, a data processing unit, and a sorting mechanism.

[0108] The optical detection module includes: a closed light box with diffuse reflection on its inner wall; a ring-shaped LED light source is provided at the top of the inner wall of the closed light box, which is an array of multi-angle lighting; an RGB industrial camera is arranged inside the closed light box for taking pictures of the fruits to be detected; a cross-polarizer is arranged between the ring-shaped LED light source and the RGB industrial camera to suppress surface highlights.

[0109] The image acquisition mechanism is specifically a turntable or a multi-camera layout structure, which is used to send the fruits to be detected into the optical detection module and cooperate with the optical detection module to achieve multi-angle image acquisition of the fruits to be detected.

[0110] The calibration module is used to calibrate the multi-angle images of the fruits to be detected. It has a built-in gray card or color card, connects to the RGB industrial camera, and automatically corrects the images when the device is powered on to achieve online white balance and color calibration.

[0111] The data processing unit, including an embedded processor or edge computing platform, is used to perform image preprocessing, composite yellowing index (FI) calculation, fruit classification algorithm, and threshold correction logic. Simultaneously, it supports communication with the cloud to receive updated parameters sent from the cloud, including weighting coefficients for calculating the composite yellowing index and / or threshold parameters for grading.

[0112] The sorting mechanism is a three-channel diversion device, including: a healthy fruit screening channel, a suspicious fruit screening channel, and a water-core disease removal channel.

[0113] In one embodiment, the suspicious fruit screening channel of the sorting mechanism is connected to a secondary detection module or a manual verification port.

[0114] In this embodiment, a non-destructive testing device for water core disease in green pears specifically includes the following components.

[0115] 1. Optical inspection module.

[0116] Enclosed light box with diffuse reflection on the inner wall.

[0117] Ring-shaped LED light source, multi-angle lighting array, and supports time-division lighting, CRI≥95, color temperature 5000–5500K.

[0118] A cross polarizer is placed between the light source and the camera to suppress surface highlights.

[0119] RGB industrial camera, resolution ≥2MP, frame rate ≥30fps.

[0120] 2. Data collection agency.

[0121] Circular track multi-angle acquisition mechanism.

[0122] 3. Calibration module.

[0123] With a built-in gray or color card, the device automatically acquires and corrects images upon startup, enabling online white balance and color calibration.

[0124] 4. Data processing unit.

[0125] Embedded processors or edge computing platforms (ARM / GPU) run image preprocessing, composite yellowing index calculation, grading algorithms, and threshold correction logic.

[0126] It supports communication with the cloud for model updates. In this embodiment, model updates refer to maintaining model stability across different batches by adaptively adjusting preset thresholds and weighting coefficients used to calculate the composite yellowing index without changing the overall detection method.

[0127] 5. Sorting mechanism.

[0128] The three-channel diversion device includes: a healthy fruit channel, a suspected fruit verification channel, and a water-core disease removal channel.

[0129] The suspicious result channel can be connected to a secondary detection module (such as single-band near-infrared detection) or a manual verification port.

[0130] The operating procedure of the device is as follows: Place the green pear on the fruit moving platform 3, and move it through the light-blocking curtain 4-2 into the RGB detection dark box for identification and detection by the industrial camera 4-6. If re-inspection is required, it passes through the near-infrared re-inspection dark box 5 and is detected by the near-infrared camera 5-1. After identification and grading, when it moves to the electric push rod 1 of the corresponding grade, the push rod responds and pushes the pear into the sorting slot 2 of the corresponding grade.

[0131] Instructions for use of the RGB detection dark box: The RGB detection dark box 4 can also be used as an image acquisition chamber. Whether for image acquisition or detection, it should first be calibrated by aligning it with the calibration color chart 4-7 using the circular track 4-4 and the multi-degree-of-freedom camera joystick 4-5 before use. The combination of the circular track 4-4 and the multi-degree-of-freedom camera joystick 4-5 facilitates the acquisition of multi-angle images of fruits and the identification of pears from multiple angles.

[0132] In one embodiment, a single-stage detection and sorting device based on the RGB composite yellowing index is provided.

[0133] like Figures 5-7 As shown, the detection device in this embodiment mainly includes a light box, a ring LED light source and a cross polarizer, an RGB camera, a turntable, a data processing unit, and a three-channel sorting mechanism. Fruits enter the light box via a conveyor belt, and a circular track multi-angle acquisition mechanism acquires images of the fruit from multiple angles, ensuring that the fruit's skin is imaged from multiple angles. The ring LED light source provides constant color temperature illumination, and the polarizer is used to suppress highlights on the skin. After acquiring fruit images, the RGB camera transmits them to the data processing unit.

[0134] like Figure 1 As shown, the data processing unit performs the following steps on the acquired images in sequence: white balance and color correction → fruit region segmentation → feature extraction → composite yellowing index calculation → three-level grading discrimination → result fusion and sorting output.

[0135] like Figure 2 As shown, the algorithm's logical framework includes a feature input module, a composite yellowing index calculation module, a three-level classification module, a multi-angle result fusion module, and an output level module. The composite yellowing index FI is calculated from color features (…). , , ), proportion of yellowed area and the texture uniformity is calculated, and the formula is as follows.

[0136] .

[0137] Set thresholds T1 and T2 to achieve a three-way discrimination of healthy fruits (FI < T1), suspicious fruits (T1 ≤ FI ≤ T2), and core rot fruits (FI > T2). After the results are output, drive the three-channel sorting mechanism to complete automatic diversion.

[0138] In one embodiment, a non-destructive detection system for fruit core rot is provided, including: a first-stage detection unit and a second-stage detection unit.

[0139] The first-stage detection unit is used to perform the detection method provided in the above embodiment to preliminarily detect and classify fruits.

[0140] The second-stage detection unit performs single-band near-infrared detection or manual review on the suspicious fruits obtained by the first-stage detection unit; wherein, the system performs final classification on the suspicious fruits according to the review results of the second detection unit.

[0141] In one embodiment, it further includes: an edge computing node and a cloud service platform.

[0142] The edge computing node is deployed at the detection site, and the first-stage detection unit is integrated on the edge computing node to implement real-time calculation of the detection method.

[0143] The cloud service platform is communicatively connected to the edge computing node, and is used to receive and store detection data, perform batch difference analysis and update parameters, and send the updated parameters to the edge computing node; wherein, the updated parameters include weight coefficients for calculating the composite chlorosis index and / or threshold parameters for grading discrimination.

[0144] Based on the method and device, the present application also proposes a two-stage detection and sorting system.

[0145] First-stage detection: Use the RGB composite chlorosis index method for rapid screening to achieve large-scale low-cost detection.

[0146] Second-stage detection: Trigger single-band near-infrared detection or manual review on suspicious fruits to improve overall accuracy.

[0147] The system realizes real-time discrimination through the edge computing unit and uploads the detection data to the cloud for batch difference analysis and model optimization. In this embodiment, model update refers to making the model maintain stability in different batches by adaptively correcting the preset thresholds and weight coefficients for calculating the composite chlorosis index without changing the overall detection method.

[0148] In one embodiment, an RGB + near-infrared two-stage detection system is provided.

[0149] Based on the single-stage detection and sorting device based on the RGB composite yellowing index in the previous embodiment, this embodiment further expands it into a two-stage detection system. First, the fruit is still sorted according to... Figure 1 and Figure 2 The process shown completes the first-level detection of the RGB composite yellowing index. For samples identified as "suspicious fruit," the system diverts them to the second-level detection channel.

[0150] In the secondary detection stage, suspected fruits are irradiated with a single-band near-infrared light source, and the transmitted / reflected light signals are collected by a near-infrared sensor or camera. The data processing unit combines the RGB results from the primary stage with the near-infrared detection results from the secondary stage for cross-validation to further confirm whether the fruit truly has watercore. If uncertainty still exists, the fruit proceeds to the manual review stage.

[0151] Through this two-level detection architecture, the system can significantly improve the overall accuracy of discrimination while ensuring high efficiency and low cost of the first-level detection.

[0152] In one embodiment, a system intelligent application is provided.

[0153] In either the single-stage detection and sorting device based on the RGB composite yellowing index or the two-stage detection system of RGB + near-infrared, as described in the foregoing embodiments, the detection data of this application can be processed in real time by the edge computing unit and uploaded to the cloud server. The cloud performs statistical analysis on data from different batches and orchards, dynamically adjusts the discrimination threshold parameters, and periodically sends data to the edge computing unit to achieve system self-learning and optimization. Furthermore, the method of this application is not only applicable to green-skinned pears but can also be extended to the detection of watercore or similar internal diseases in other pear varieties, as well as apples, citrus fruits, and other fruits.

[0154] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

Claims

1. A non-destructive method of detecting internal fruit disorders, characterized in that, The method comprises the following steps: Step S1, obtaining multiple images of the fruit to be detected, and performing image preprocessing and correction on each image; wherein the multiple images record multiple angles of the fruit to be detected; Step S2, extracting color features, area features and texture features of the fruit to be detected in the effective peel area of each image; Step S3, obtaining the fruit variety of the fruit to be detected, calling the pre-stored weight coefficient and threshold parameter for grading discrimination, and based on the color features, area features, texture features of the fruit to be detected and the weight coefficient, a composite yellowing index FI is constructed to comprehensively reflect the yellowing degree and tissue uniformity of the fruit to be detected; in the step S3, the composite yellowing index FI is specifically: ； wherein, , , denotes the weight coefficient calibrated by the training sample; , , denotes the color feature of the extracted fruit to be detected, wherein, , denotes the CIELab component, denotes the HSV hue component; denotes the area feature of the extracted fruit to be detected, i.e. the yellowing area proportion; denotes the texture feature of the extracted fruit to be detected, i.e. the texture uniformity index; Step S4, using the constructed composite yellowing index FI to calculate the yellowing index FI of the fruit to be detected in each image, and comparing each image with the preset threshold value, so as to classify each image of the fruit to be detected as healthy fruit, suspicious fruit or water core disease fruit; Step S5, fusing the calculation results of each image of the fruit to be detected, and outputting the preliminary classification result of the fruit to be detected, which is specifically healthy fruit, suspicious fruit or water core disease fruit.

2. The non-destructive testing method of claim 1, wherein, The step S4 specifically comprises: Comparing the calculated composite yellowing index FI with the preset threshold value to realize the three classifications of the fruit to be detected: Healthy fruit: FI < T1; Suspicious fruit: T1 ≤ FI ≤ T2; Water core disease fruit: FI > T2; Wherein, the first preset threshold T1 and the second preset threshold T2 are obtained by experimental data training, and are adaptively corrected according to batch difference, and the dynamic correction rule of the threshold value is: ； wherein, represents the current batch health mean, represents the reference mean, represents the correction factor; represents the dynamic correction threshold; represents the base threshold.

3. The non-destructive testing method of claim 1, wherein, The step S1 specifically comprises: using an RGB industrial camera to collect multiple angle images of the fruit to be detected; Each of the collected multiple images is subjected to white balance processing and color correction, and the image is converted to CIELab and HSV color space through standard gray card or color card calibration.

4. The non-destructive testing method of claim 1, wherein, The step S5 specifically comprises: fusing the calculation results of each image of the fruit to be detected by voting or probability weighting, and the preliminary classification result of the fruit to be detected.

5. The non-destructive testing method of claim 1, wherein, After the step S5, it further comprises: for the fruit to be detected with the preliminary classification result of suspicious fruit, performing single-band near-infrared detection or manual review to obtain the final classification result.

6. A non-destructive testing device for internal water core disease of fruit, said device being used to accomplish the testing method of claim 1, characterized in that, The method comprises the following steps: An optical detection module comprises: a closed light box with a diffuse reflection inner wall; a ring-shaped LED light source is arranged at the top of the inner wall of the closed light box, which is a multi-angle lighting array; an RGB industrial camera is arranged in the closed light box for shooting the fruit to be detected; a cross polarizer is arranged between the ring-shaped LED light source and the RGB industrial camera to suppress surface highlights; An image collection mechanism, specifically a turntable or a multi-camera arrangement structure, is used to send the fruit to be detected into the optical detection module, and cooperates with the optical detection module to realize multi-angle image collection of the fruit to be detected; A calibration module is arranged to calibrate the multi-angle images of the fruit to be detected, and is internally provided with a gray card or a color card and connected to the RGB industrial camera, so as to automatically correct the images when the device is powered on, thereby realizing online white balance and color calibration. A data processing unit is arranged to include an embedded processor or an edge computing platform, and is used to execute image preprocessing, composite yellowing index FI calculation, fruit classification algorithm and threshold correction logic. Meanwhile, the data processing unit supports communication with the cloud, and is used to receive update parameters issued by the cloud, wherein the update parameters include weight coefficients for calculating the composite yellowing index and / or threshold parameters for hierarchical discrimination. The sorting mechanism is specifically a three-channel shunting device, which includes a healthy fruit screening channel, a suspicious fruit screening channel and a water core disease fruit removing channel.

7. The non-destructive testing apparatus of claim 6, wherein, The suspicious fruit screening channel of the sorting mechanism is connected to the secondary detection module or the manual review port.

8. A non-destructive testing system for internal fruit quality, characterized in that, The system comprises: A first-level detection unit is arranged to execute the detection method of claim 1 to preliminarily detect and classify the fruit; A second-level detection unit is arranged to perform single-band near-infrared detection or manual review on the suspicious fruit obtained by the first-level detection unit; The system classifies the suspicious fruit according to the review result of the second-level detection unit.

9. The non-destructive testing system of claim 8, wherein, Further comprising: An edge computing node is arranged at the detection site, and the first-level detection unit is integrated on the edge computing node to realize real-time calculation of the detection method; A cloud service platform is in communication connection with the edge computing node, and is used to receive and store detection data, perform batch difference analysis and update parameters, and issue the update parameters to the edge computing node; wherein the update parameters include weight coefficients for calculating the composite yellowing index and / or threshold parameters for hierarchical discrimination.

Images