Fruit tray and fruit surface defect recognition device

Abstract

The invention provides a fruit tray and a fruit surface defect recognition device. The fruit surface defect recognition device comprises a conveyor belt; two bottom spiral guide rails of the fruit tray are in contact with the conveying belt through the bottoms to obtain friction with the conveying belt to form self-rotation, and are in contact with the fruits through the tops to drive the fruits to rotate and roll over; the multi-node top camera system is arranged right above the conveying belt, each node is provided with a wide-angle lens and a classification model, local images of the upper half face of the fruit to be detected right below the wide-angle lens are shot through the wide-angle lens, and the local images are classified in real time through the classification model so as to output a detection result used for identifying whether the fruit to be detected has flaws or not; and the dynamic sorting control system is used for triggering a sorting action according to the detection result of each node. By means of the fruit defect detection device, it can be ensured that fruits are evenly overturned in the conveying process, a camera can shoot the surfaces of the fruits from different angles, and then surface full-covering fruit defect detection is achieved.

Description

Technical Field

[0001] The present application relates to the field of machine automation detection, and specifically, to a device for identifying surface defects of fruit plates and fruits. Background Art

[0002] In the field of fruit processing and sorting, surface defect detection is a critical step, directly impacting the quality and market acceptance of the final product. Traditional methods for fruit surface defect detection rely primarily on manual visual inspection or simple mechanical devices. These methods are not only inefficient but also susceptible to human factors, leading to inconsistent detection results and high false positive rates. With the development of automation and image processing technologies, more and more automated inspection systems are being introduced into the fruit sorting process.

[0003] However, existing automated fruit surface defect detection systems still have some shortcomings. First, many systems use a single-view camera to capture images of the fruit surface. This method often cannot fully cover the entire surface of the fruit, especially for fruits with irregular shapes or complex surface features, which can easily lead to detection blind spots. Secondly, although some systems use multi-view or multi-camera solutions, these solutions usually require complex mechanical structures to achieve the rotation and tumbling of the fruit, which not only increases the complexity and cost of the system, but may also lead to an increase in equipment failure rates. In addition, some existing solutions rely on high-performance computing resources for image processing, which may become a bottleneck on large-scale production lines, affecting overall efficiency.

[0004] Therefore, the present application provides a fruit plate and fruit surface defect identification device to solve one of the above technical problems. Summary of the Invention

[0005] The purpose of this application is to provide a device for identifying surface defects of fruit plates and fruits, which can solve at least one of the technical problems mentioned above. The specific solution is as follows: According to a specific embodiment of the present application, in a first aspect, the present application provides a fruit plate, comprising: Two bottom spiral guide rails, the spiral guide rails are cylindrical in configuration, and the cylindrical side surfaces of the two bottom spiral guide rails are set to have spiral textures with the same rotation direction; a first connecting plate is respectively connected to the first end heads of the two bottom spiral guide rails; a second connecting plate is respectively connected to the second end heads of the two bottom spiral guide rails; wherein the two bottom spiral guide rails contact the conveyor belt through the bottom to obtain friction with the conveyor belt to form self-rotation, and contact the fruit through the top to drive the rotation and tumbling of the fruit.

[0006] According to a specific embodiment of the present application, in a second aspect, the present application provides a device for identifying surface defects of fruit, comprising: A conveyor belt; the fruit tray described in the first aspect; a multi-node top camera system arranged directly above the conveyor belt, each node being equipped with a wide-angle lens and a classification model, for capturing a partial image of the upper half of the fruit to be inspected directly below the conveyor belt through the wide-angle lens, and for performing real-time classification of the partial image through the classification model to output a detection result for identifying whether the fruit to be inspected has defects; and a dynamic sorting control system for triggering a sorting action based on the detection results of each node.

[0007] In one embodiment, the dynamic sorting control system triggers the sorting action in the following manner: if a defect is detected through the first node, the sorting action is directly triggered; if no defect is detected through the first node, defect detection is performed through subsequent nodes of the first node, and when a defect is detected at any of the subsequent nodes, a defect area assessment based on defect area fitting is performed, so as to trigger the sorting action when the defect area reaches an area threshold; wherein the calculation formula for the defect area assessment is: ;in, is the defect area after fitting, is the defect area detected at the i-th node, The coverage area weight of the node; the coverage area weight is set based on the fruit tray spacing and the spiral guide rail speed, and the setting range is 0.5 ≤ ≤ 0.7.

[0008] In one embodiment, the classification model adopts the YOLO v5 architecture; the local image is an RGB three-channel image with a resolution of 128×128 pixels; The detection results are defect category labels, including "no defect", "dent", "rot" and "worm-eaten". The classification model is set to output detection results that meet a confidence threshold after pre-training, and the confidence threshold is 0.8.

[0009] In one embodiment, the dynamic sorting control system is configured with a redundant verification mechanism for verifying the detection results of each node before triggering the sorting action; the redundant verification mechanism adopts a multi-node data cross-validation method and calculates the comprehensive confidence of the detection results between multiple nodes using the following formula: ;in, is the comprehensive confidence of the detection results among multiple nodes, is the confidence of the detection result of the jth node, is the confidence weight of the jth node; A value ≥ 0.9 indicates a defect. The confidence level of the test result is dynamically adjusted based on the defect category label of the test result. The confidence weight of "rot" is 1.2, the confidence weight of "worm-eaten" is 0.8, and the confidence weight of "sag" is 0.9.

[0010] In one embodiment, the parametric design of the spiral guide rail of the fruit tray meets the following conditions: the spacing is between 20 cm and 40 cm, and the inclination angle is between 0.4 rad and 0.6 rad; the friction coefficient between the top of the spiral guide rail and the contact surface of the fruit is dynamically adjusted according to the characteristics of the fruit skin, the friction coefficient of apples is 0.3 to 0.5, and the friction coefficient of citrus fruits is 0.1 to 0.2.

[0011] In one embodiment, the wide-angle lens coverage angle of each node in the multi-node top camera system is 180° to 220°, and the shooting areas of adjacent nodes are set to have an overlapping area of 10% to 30%.

[0012] In one embodiment, the device supports modular expansion, and each channel is independently configured with a spiral fruit plate, a top camera and an edge computing unit; wherein the edge computing unit is deployed locally on the node, processes images in real time and transmits the results, reducing the load on the central server.

[0013] In one embodiment, the device further includes a dynamic parameter adaptive adjustment module for automatically adjusting the spiral guide rail parameters and camera parameters according to the fruit type or surface characteristics; when the fruit type is apple, the dynamic parameter adaptive adjustment module sets the spacing of the spiral guide rail to 30 cm, the tilt angle to 0.5 rad, and the exposure time of the wide-angle lens to 1 / 1000 second; when the fruit type is orange, the dynamic parameter adaptive adjustment module sets the spacing of the spiral guide rail to 35 cm, the tilt angle to 0.45 rad, and the exposure time of the wide-angle lens to 1 / 500 second.

[0014] In one embodiment, the dynamic parameter adaptive adjustment module calculates the fruit rotation angle using the following formula: ;in, is the fruit rotation angle, L is the transmission path length, r is the fruit radius, and α is the spiral guide rail inclination angle; the dynamic parameter adaptive adjustment module adjusts the spiral guide rail parameters and the camera parameters based on the standard of θ≥360° so that the fruit to be inspected meets the detection requirement of full surface coverage.

[0015] Compared with the prior art, the above-mentioned scheme of the embodiment of the present application has at least the following beneficial effects: the fruit surface defect identification device provided by the present application includes a conveyor belt, a fruit tray, and a multi-node top camera system. Each node is equipped with a wide-angle lens and a classification model. The wide-angle lens captures a partial image of the upper half of the fruit to be inspected directly below, and these images are classified in real time using the classification model. The detection results used to identify whether the fruit has defects are output, and the dynamic sorting control system triggers the corresponding sorting action based on the detection results of each node. This design not only achieves multi-angle coverage of the fruit surface, but also significantly improves detection efficiency through real-time classification and rapid response mechanisms, reduces the error rate, and ensures the efficient operation of the production line. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 Shown is a structural schematic diagram of a fruit plate; Figure 2 The figure shows a structural diagram of a device for identifying fruit surface defects. DETAILED DESCRIPTION

[0017] To make the objectives, technical solutions, and advantages of this application more clear, this application will be further described in detail below with reference to the accompanying drawings. Obviously, the embodiments described are only some of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making any creative efforts are within the scope of protection of this application.

[0018] The terms used in the examples of this application are for the purpose of describing specific embodiments only and are not intended to limit this application. The singular forms "a," "the," and "the" used in the examples of this application and the appended claims are also intended to include plural forms, and unless the context clearly indicates otherwise, "a plurality" generally includes at least two.

[0019] It should be understood that the term "and / or" as used herein is merely a description of the relationship between associated objects, indicating that three possible relationships exist. For example, "A and / or B" can represent: A exists alone, A and B exist simultaneously, or B exists alone. Furthermore, the character " / " in this document generally indicates that the associated objects are in an "or" relationship.

[0020] It should be understood that although the terms first, second, third, etc. may be used to describe in the embodiments of the present application, these descriptions should not be limited to these terms. These terms are only used to distinguish the descriptions. For example, without departing from the scope of the embodiments of the present application, the first may also be referred to as the second, and similarly, the second may also be referred to as the first.

[0021] As used herein, the words "if" and "if" may be interpreted as "at the time of" or "when" or "in response to determining" or "in response to detecting," depending on the context. Similarly, the phrases "if it is determined" or "if (stated condition or event) is detected" may be interpreted as "when it is determined" or "in response to the determination" or "when detecting (stated condition or event)" or "in response to detecting (stated condition or event)," depending on the context.

[0022] It should also be noted that the terms "include," "comprises," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a product or device comprising a series of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such product or device. In the absence of further limitations, an element defined by the phrase "comprises a..." does not exclude the presence of other identical elements in the product or device comprising the element.

[0023] It should be noted in particular that any symbols and / or numbers in the specification that are not marked in the accompanying drawings are not drawing marks.

[0024] The optional embodiments of the present application are described in detail below with reference to the accompanying drawings.

[0025] The following combination Figure 1 The embodiments of the present application are described in detail.

[0026] Figure 1 A structural diagram of a fruit plate 11 is shown. Figure 1 As shown, the fruit tray 11 includes two bottom spiral guide rails 110 , a first connecting plate 111 and a second connecting plate 112 .

[0027] For example, the two bottom spiral guide rails 110 are cylindrical in configuration, and the cylindrical side surfaces of the two bottom spiral guide rails 110 are configured to have spiral textures with the same hand direction.

[0028] For example, the first connecting plate 111 is connected to the first ends of the two bottom spiral guide rails 110, and the second connecting plate 112 is connected to the second ends of the two bottom spiral guide rails 110. The term "end" refers to a circular surface portion of a cylindrical configuration, for example, one side of the circular surface represents the first end, and the other side of the circular surface represents the second end.

[0029] The two bottom spiral guide rails 110 contact the conveyor belt 10 at the bottom to obtain friction with the conveyor belt 10 to form self-rotation, and contact the fruit at the top to drive the fruit to rotate and roll.

[0030] In the embodiment of the present application, the fruit tray 11 is designed with two bottom spiral guide rails 110, which have the same rotation direction and spiral texture on the cylindrical side surface. The two spiral guide rails are fixed together by a first connecting plate 111 and a second connecting plate 112 to form a stable structure. When the fruit tray 11 moves on the conveyor belt, the bottom spiral guide rail 110 contacts the conveyor belt and rotates by friction, thereby driving the fruit on the top to rotate and roll. This design avoids complex mechanical rotation devices, simplifies the equipment structure, reduces costs, and can ensure that the fruit is evenly flipped during transportation, allowing the camera to capture the surface of the fruit from different angles, improving the comprehensiveness and accuracy of detection.

[0031] The present application also provides an apparatus embodiment that is consistent with the above embodiment, which is used to express the actual function of the fruit plate 11 in the above embodiment. The explanation based on the same name meaning is the same as that in the above embodiment, and has the same technical effect as the above embodiment, which will not be repeated here.

[0032] Figure 2 The figure shows a structural diagram of a device for identifying fruit surface defects.

[0033] For example, Figure 2 As shown, the fruit surface defect recognition device 1 includes a conveyor belt 10, a fruit tray 11, a multi-node top camera system 12 and a dynamic sorting control system 13, and in some embodiments further includes a dynamic parameter adaptive adjustment module 14.

[0034] For example, the fruit tray 11 includes two bottom spiral guide rails 110, a first connecting plate 111, and a second connecting plate 112. The two bottom spiral guide rails 110 are cylindrical, and the cylindrical side surfaces of the two bottom spiral guide rails 110 are provided with spiral textures having the same hand direction. The first connecting plate 111 is respectively connected to the first ends of the two bottom spiral guide rails 110, and the second connecting plate 112 is respectively connected to the second ends of the two bottom spiral guide rails 110.

[0035] The multi-node top camera system 12 is arranged directly above the conveyor belt 10. Each node is equipped with a wide-angle lens 121 and a classification model 122. The wide-angle lens 121 is used to capture a local image of the fruit to be inspected, and the classification model 122 is used to classify the local image in real time to output a detection result for identifying whether the fruit to be inspected has defects.

[0036] For example, Figure 2 As shown, the spacing between the wide-angle lenses 121 of adjacent nodes in the multi-node top camera system 12 can be configured to be the same. Figure 2 Indicated by D.

[0037] For example, Figure 2As shown, the dynamic sorting control system 13 and the dynamic parameter adaptive adjustment module 14 can be deployed in a node-corresponding manner, that is, each node is respectively configured with a sub-end of the dynamic sorting control system 13 and a sub-end of the dynamic parameter adaptive adjustment module 14. Correspondingly, the wide-angle lens 121 and the classification model 122 equipped at each node also serve as the sub-ends configured at each node of the multi-node top camera system 12. On this basis, each node integrates a set of sub-ends to independently complete its own tasks, perform image acquisition, defect recognition, sorting control, and / or parameter adjustment actions, and the entire execution process adopts a one-to-one processing method. At the same time, each node can communicate and be controlled by each parent end including the multi-node top camera system 12, the dynamic sorting control system 13, and the dynamic parameter adaptive adjustment module 14, and / or communicate and collaborate with other nodes through wired or wireless means.

[0038] Furthermore, it is understood that, in addition to the node-by-node configuration described above, the dynamic sorting control system 13 and the dynamic parameter adaptive adjustment module 14 can also be configured independently as a data processing and control center. In this approach, the dynamic sorting control system 13 and / or the dynamic parameter adaptive adjustment module 14 receive data from each node, including captured images and currently configured parameters, and perform centralized processing. The entire execution process utilizes a one-to-many processing approach.

[0039] For example, the two bottom spiral guide rails 110 of the fruit tray 11 contact the conveyor belt 10 at the bottom to obtain friction with the conveyor belt 10 to form self-rotation, and contact the fruit to be inspected at the top to drive the rotation and rolling of the fruit, so that the fruit to be inspected shows the same or different surfaces when located at different nodes.

[0040] The dynamic sorting control system 13 is used to trigger sorting actions according to the detection results of each node.

[0041] In this application, the fruit surface defect identification device 1 includes a conveyor belt 10, a fruit tray 11, and a multi-node top camera system 12. Each node is equipped with a wide-angle lens 121 and a classification model 122. The wide-angle lens 121 captures partial images of the fruit to be inspected, and the classification model 122 is used to classify these images in real time, outputting a detection result for identifying whether the fruit has defects. The dynamic sorting control system 13 triggers the corresponding sorting action based on the detection results of each node. This design not only achieves multi-angle coverage of the fruit surface, but also significantly improves detection efficiency through real-time classification and rapid response mechanisms, reduces the error rate, and ensures the efficient operation of the production line.

[0042] In some embodiments, each channel can be designed as an independent module, each equipped with its own spiral fruit disc and top camera, allowing the number of channels to be increased or decreased on demand, for example, from 2-3 channels to 8 channels. This solution allows the production line to flexibly adjust to production capacity needs, adapting to different scenarios from small-scale trials to large-scale industrial production, enhancing the versatility and cost-effectiveness of the equipment. At the same time, by adjusting the spiral pitch, inclination angle, and conveyor belt parameters, it can be expanded to non-fruit applications, such as detecting surface defects in spherical industrial parts such as bearing balls, providing a highly efficient solution for production line quality control.

[0043] In some embodiments, the fruit surface defect recognition device 1 can also be used to detect the sealing of fruit packaging bags, for example, to detect whether the packaging is leaking or damaged.

[0044] For example, the dynamic sorting control system 13 triggers the sorting action as follows. If a defect is detected by the first node, the sorting action is directly triggered. If no defect is detected by the first node, defect detection is performed using subsequent nodes. Because the fruit rotates and rolls evenly on the fruit tray, each node accurately calculates and captures a random half-surface image of the fruit to be inspected directly below it. After a certain number of nodes are used, the entire surface of the fruit can be fully covered.

[0045] Furthermore, when a defect is detected at any subsequent node, a defect area assessment based on defect area fitting is performed to trigger a sorting action when the defect area reaches an area threshold.

[0046] Among them, the defect area is evaluated using the formula .in, is the defect area after fitting, is the defect area detected at the i-th node, Overrides the region weight for this node.

[0047] In some embodiments, the coverage area weight is set based on the spacing between the fruit trays 11 and the speed of the spiral guide rail, and the setting range is 0.5 ≤ ≤ 0.7.

[0048] In the embodiment of the present application, the dynamic sorting control system 13 adopts a phased triggering mechanism based on the node detection results. If the first node detects a defect, the sorting action is triggered immediately without waiting for the results of subsequent nodes, which greatly shortens the sorting delay. If the first node does not detect a defect, the subsequent nodes continue to detect. On this basis, once any subsequent node detects a defect, the system will perform a defect area assessment based on defect area fitting, and then the sorting action will be triggered only when the defect area reaches a preset threshold. This method not only ensures high-precision defect detection, but also avoids unnecessary frequent sorting actions, thereby improving the overall efficiency of the system.

[0049] In some embodiments, the classification model 122 may employ the YOLO v5 architecture.

[0050] In some embodiments, the local image is an RGB three-channel image with a resolution of 128×128 pixels.

[0051] In some embodiments, the detection result is a defect category label including “no defect”, “dent”, “rot”, and “insect-eaten”.

[0052] In some embodiments, the classification model 122 is configured to output detection results that meet a confidence threshold after pre-training, where the confidence threshold is 0.8.

[0053] In some embodiments, the dynamic sorting control system 13 is configured with a redundant verification mechanism for verifying the detection results of each node before triggering the sorting action.

[0054] Among them, the redundant verification mechanism adopts multi-node data cross-validation method, for example, it can be achieved through the formula Calculate the comprehensive confidence of the detection results among multiple nodes. is the comprehensive confidence of the detection results among multiple nodes, is the confidence of the detection result of the jth node, is the confidence weight of the j-th node.

[0055] in, A value ≥ 0.9 indicates a defect.

[0056] The confidence of the detection results is dynamically adjusted according to the defect category label of the detection results. The confidence weight of "rot" is 1.2, the confidence weight of "worm-eaten" is 0.8, and the confidence weight of "sag" is 0.9.

[0057] In this application, the dynamic sorting control system 13 is configured with a redundant verification mechanism to improve the accuracy of detection by cross-verifying the detection results of each node. Specifically, the system calculates the comprehensive confidence by the formula and A value ≥0.9 indicates a defect. The confidence level of the test results is dynamically adjusted based on the defect type. For example, the confidence level for "rot" is 1.2, the confidence level for "worm damage" is 0.8, and the confidence level for "dent" is 0.9. This multi-level verification mechanism effectively reduces the false positive rate and ensures the high reliability and accuracy of the test results.

[0058] In some embodiments, the redundant verification phase can incorporate deep learning models, such as Mask R-CNN, to quantitatively grade defect areas and locations, identifying specific categories such as "minor scratches" or "5% rot." This design supports the subdivision of sorting grades, classifying fruit into multiple categories such as "defective," "downgraded," and "qualified," thereby increasing commercial value. It is suitable for scenarios requiring higher sorting accuracy. Software upgrades can also expand this to agricultural product grading, adding shape, size, and color analysis capabilities. It can also be combined with robotic arms or conveyor belt branches to achieve automated packaging, further improving the efficiency of the agricultural supply chain.

[0059] In some embodiments, the parametric design of the spiral guide rail of the driving mechanism of the spiral fruit tray 11 satisfies the following conditions: Condition 1: The spacing is between 20cm and 40cm.

[0060] Condition 2: The tilt angle is 0.4 rad to 0.6 rad.

[0061] Condition three: The friction coefficient between the bottom of the fruit plate 11 and the contact surface of the fruit is dynamically adjusted according to the characteristics of the fruit skin.

[0062] In some preferred embodiments, the coefficient of friction of apples is 0.3 to 0.5, and the coefficient of friction of citrus fruits is 0.1 to 0.2.

[0063] In the embodiment of the present application, the parametric design of the spiral guide rail of the driving mechanism of the spiral fruit tray 11 meets specific conditions, ensuring that different types of fruits can be evenly flipped during the transmission process, avoiding detection blind spots or jamming caused by insufficient or excessive friction, thereby improving the comprehensiveness and stability of detection.

[0064] In some embodiments, the wide-angle lens 121 of each node in the multi-node top camera system 12 covers a viewing angle of 180° to 220°, and the image capture areas of adjacent nodes overlap by 10% to 30%. This design ensures comprehensive coverage of the fruit surface; even if a single node has blind spots within its field of view, the overlapping areas of adjacent nodes can compensate for these shortcomings.

[0065] In some embodiments, the device supports modular expansion, with each channel independently configured with a spiral fruit tray 11, a top camera, and an edge computing unit. The edge computing unit is deployed locally on the node, processing images and transmitting the results in real time, reducing the load on the central server. This modular design not only improves the system's flexibility and scalability, but also accelerates image processing through localized processing, reduces network latency, and ensures real-time and high efficiency, making it particularly suitable for large-scale production line applications.

[0066] In some embodiments, the device further includes a dynamic parameter adaptive adjustment module 14, which is used to automatically adjust the spiral guide rail parameters and camera parameters according to the fruit type or surface characteristics.

[0067] In some preferred embodiments, when the fruit type is apple, the dynamic parameter adaptive adjustment module 14 sets the pitch of the spiral guide rail to 30 cm, the tilt angle to 0.5 rad, and the exposure time of the wide-angle lens 121 to 1 / 1000 second.

[0068] In some preferred embodiments, when the fruit type is orange, the dynamic parameter adaptive adjustment module 14 sets the pitch of the spiral guide rail to 35 cm, the tilt angle to 0.45 rad, and the exposure time of the wide-angle lens 121 to 1 / 500 second.

[0069] In the embodiments of the present application, this intelligent adjustment mechanism optimizes detection parameters based on the characteristics of different fruits, ensuring optimal detection results while simplifying the operational process and improving the system's applicability and ease of use. The adjustment of the spacing and tilt angle is optimized based on considerations of friction between the fruit and the guide rail and transmission efficiency, while the exposure time is optimized based on considerations of the fruit's surface reflective properties and size. These parameters ensure clear, fully covered images of the fruit as it passes through the device.

[0070] In some embodiments, the dynamic parameter adaptive adjustment module 14 calculates the fruit rotation angle using the following formula: .

[0071] in, is the fruit rotation angle, L is the length of the conveying path, r is the fruit radius, and α is the inclination angle of the spiral guide.

[0072] The dynamic parameter adaptive adjustment module 14 adjusts the spiral guide rail parameters and the camera parameters based on the standard of θ≥360°, so that the fruit to be inspected meets the inspection requirement of full surface coverage.

[0073] For example, as fruit passes through conveyor belt 10 and is imaged by a camera for defect detection, the system intelligently adjusts the drive mode of spiral fruit tray 11 based on the fruit's size and motion to ensure that the fruit is captured from all angles during transport and to avoid visual blind spots. Assume that an apple, with an average diameter of approximately 7 cm, is being inspected by the device. The node spacing of conveyor belt 10 is set to 30 cm, and the inclination angle of the spiral guide rail is set to 0.5 rad. The above formula yields: Approximately 46.8°. Clearly, the fruit has only rotated approximately 46.8°, far from achieving the required 360° coverage. Therefore, the dynamic parameter adaptive adjustment module 14 automatically identifies the current fruit type and, based on preset parameter matching rules, gradually optimizes the geometric parameters of the spiral guide rail.

[0074] For example, if, after a new round of parameter adjustments, the spiral guideway's inclination angle is increased to 0.6 rad and the spiral pitch is reduced from 7.5 cm to 5 cm, the fruit will experience a higher tumbling frequency and a wider viewing angle range within the same 30 cm conveyor path. This approach allows for multi-angle image information of the fruit at different time points, improving the completeness and effectiveness of image acquisition.

[0075] Furthermore, the module synchronizes the camera sampling frequency with the actual speed of the fruit. For example, if the current linear speed of conveyor belt 10 is 28 cm / s and the camera frame rate is 60 frames per second, the system ensures that at least one clear image is captured for each critical rotation angle, avoiding missed detections due to image blur or jumps. This dynamic collaborative control mechanism not only improves detection coverage but also enhances the system's robustness and adaptability. Furthermore, when the inspection object is replaced with a larger, smooth-skinned tomato, which has poor rolling stability and tends to slide rather than roll, the dynamic parameter adaptive adjustment module 14 adjusts the friction coefficient between the bottom of the fruit tray 11 and the fruit to maintain it between 0.1 and 0.2, ensuring uniform rotation of the tomato under the action of the spiral guide. Simultaneously, the camera exposure time is fine-tuned based on the reflective properties of the fruit, for example, from 1 / 1000 second to 1 / 500 second, to reduce overexposure in highlight areas and improve image quality. In this way, the dynamic parameter adaptive adjustment module 14 provided in this application can automatically optimize the mechanical structure parameters and image acquisition strategies under different fruit types, different production rhythms and different lighting conditions, ensuring that each inspection has complete surface coverage capabilities and stable image quality, thereby significantly improving the accuracy and applicability of the overall inspection system.

[0076] In some embodiments, the present application provides a fruit surface defect identification device 1, designed to detect surface defects on round fruits (such as apples, tomatoes, and grapefruits) on a high-speed conveyor belt 10. Prior art single-camera plus rotating mechanism solutions are limited by mechanical response speed and struggle to meet high-speed sorting requirements. While multi-camera systems can improve detection efficiency, they suffer from high hardware costs and complex algorithms. Single-camera, wide-area capture solutions, however, suffer from high error rates due to difficulties in image segmentation and matching. To address this, the present application proposes a sorting solution based on a collaborative design of a spiral fruit tray 11 and a multi-node top camera, significantly reducing equipment size and operating costs while ensuring detection coverage.

[0077] As a feasible embodiment, a fruit surface defect identification device 1 includes a conveyor belt 10, a spiral fruit tray 11, a multi-node top camera system 12, and a dynamic sorting control system 13. The bottom of the spiral fruit tray 11 is equipped with two spiral guide rails with the same rotation direction, with the two ends fixed by a first connecting plate 111 and a second connecting plate 112. As the fruit tray 11 moves on the conveyor belt 10, the bottom spiral guide rails 110 contact the conveyor belt 10 and drive the fruit tray 11 to rotate through friction, thereby causing the fruit to rotate and tumble. This design eliminates the need for complex motor drive structures and allows the fruit to naturally flip during transportation, providing sufficient viewing angle changes for subsequent image acquisition.

[0078] As an extended embodiment, the device supports modular expansion, and each channel is independently configured with a spiral fruit plate 11, a top camera, and an edge computing unit. The edge computing unit is deployed locally at the node, processes images in real time, and transmits the results, reducing the load on the central server and improving the system response speed and stability. In addition, the device also includes a dynamic parameter adaptive adjustment module 14, which can automatically adjust the spiral guide rail parameters and camera parameters according to the type of fruit or surface characteristics. For example, for apple fruits, the dynamic parameter adaptive adjustment module 14 sets the spiral guide rail spacing to 30 cm, the tilt angle to 0.5 rad, and the exposure time of the wide-angle lens 121 to 1 / 1000 second. For orange fruits, it is set to 35 cm spacing, 0.45 rad tilt angle, and 1 / 500 second exposure time.

[0079] In some embodiments, fruit motion control and detection logic is combined with spiral guide rail parameters to precisely match the speed of conveyor belt 10. For example, with a conveyor belt 10 speed of 28 cm / s, a spiral guide rail spacing of 7.5 cm, and a tilt angle of 0.5 rad, apples can complete a rotation of approximately 490° and a roll of 180° within a 30 cm node spacing, ensuring that the camera nodes can capture complete information about the fruit surface. Using two nodes for camera capture, the total coverage angle reaches 294.59°, although some blind spots still exist. Using three nodes for camera capture, the total coverage angle reaches 409.18°, completely covering the fruit surface.

[0080] As a typical application scenario, the software's sorting logic adopts multi-channel independent judgment and phased dynamic decision-making process. For multi-channel design, the fruit in each channel is divided into independent areas from a top perspective. The software only analyzes the local image of the fruit in the current channel, such as the left half or the right half, without associating with other channel data. For example, node 1 runs classification model 122 on the left half image of channel 1 and outputs a "defective" or "no defect" label. If node 1 detects a defect, the sorting action is triggered immediately to avoid waiting for the results of subsequent nodes, significantly reducing sorting delays. If node 1 does not detect a defect, node 2 continues to analyze the right half image. If a new defect is found, the defect area is evaluated by combining the data of the two nodes to further improve the accuracy of the judgment.

[0081] In summary, this application uses a spiral fruit tray to drive the rotation and tumbling of fruit, combined with a multi-node top camera system and a lightweight classification model 122, to achieve efficient, low-cost, and high-coverage fruit surface defect detection under high-speed, multi-channel conditions. This solution not only simplifies the mechanical structure and algorithm logic, but also improves detection accuracy and sorting efficiency, making it particularly suitable for application scenarios in industrial large-scale production lines.

[0082] Although operations are described in a particular order in the drawings, this should not be understood as requiring that the operations be performed in the particular order shown or in serial order, or that all shown operations be performed to obtain the desired results. In certain circumstances, multitasking and parallel processing may be advantageous.

[0083] Any steps, operations or procedures described herein may be performed or implemented using one or more hardware or software modules, either alone or in combination with other devices. In one embodiment, the software modules are implemented using a computer program product comprising a computer-readable medium containing computer program code, which can be executed by a computer processor to perform any or all of the steps, operations or procedures described.

[0084] The foregoing description of the implementation of the present application has been provided for purposes of illustration and description. The foregoing description is not intended to be exhaustive or to limit the present application to the precise form disclosed, and various variations and modifications are possible in accordance with the above teachings or may result from the practice of the present application. These embodiments have been selected and described in order to illustrate the principles of the present application and its practical application, so as to enable those skilled in the art to utilize the present application in various embodiments and modifications as appropriate for the particular use contemplated.

[0085] It is further understood that, unless otherwise specified, “connection” includes a direct connection where there are no other components between the two elements, and also includes an indirect connection where there are other elements between the two elements.

[0086] It should be further understood that although operations are described in a particular order in the drawings in the embodiments of the present application, this should not be construed as requiring that these operations be performed in the particular order shown or in a serial order, or that all of the illustrated operations be performed to obtain the desired result. In certain circumstances, multitasking and parallel processing may be advantageous.

[0087] Those skilled in the art will readily appreciate other embodiments of the present application after considering the specification and practicing the invention disclosed herein. This application is intended to encompass any variations, uses, or adaptations of the present application that follow the general principles of the present application and include common knowledge or customary techniques in the field of the present application that are not disclosed herein. The specification and examples are intended to be exemplary only, and the true scope and spirit of the present application are indicated by the scope of claims below.

[0088] It should be understood that the present application is not limited to the precise structures described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present application is limited only by the scope of the appended claims.

[0089] The above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A fruit plate, characterized in that: include: Two bottom spiral guide rails, each of which is cylindrical, and the cylindrical side surfaces of the two bottom spiral guide rails are provided with spiral textures having the same hand direction; a first connecting plate, connected to the first ends of the two bottom spiral guide rails respectively; a second connecting plate, connected to the second ends of the two bottom spiral guide rails respectively; The two bottom spiral guide rails contact the conveyor belt at the bottom to obtain friction with the conveyor belt to form self-rotation, and contact the fruit at the top to drive the fruit to rotate and roll.

2. A fruit surface defect recognition device, characterized in that: include: conveyor belts; The fruit plate according to claim 1; A multi-node top camera system is arranged directly above the conveyor belt, with each node equipped with a wide-angle lens and a classification model. The wide-angle lens captures a partial image of the upper half of the fruit to be inspected directly below, and the classification model performs real-time classification on the partial image to output a detection result indicating whether the fruit to be inspected has defects. A dynamic sorting control system is used to trigger sorting actions according to the detection results of each of the nodes.

3. The device according to claim 2, characterized in that The dynamic sorting control system triggers the sorting action in the following manner, including: If a defect is detected by the first node, the sorting action is triggered directly; If no defect is detected at the first node, defect detection is performed at subsequent nodes of the first node. When a defect is detected at any of the subsequent nodes, a defect area assessment based on defect area fitting is performed, so as to trigger a sorting action when the defect area reaches an area threshold. The calculation formula for defect area assessment is: ； in, is the defect area after fitting, is the defect area detected at the i-th node, Covering area weight for this node; The coverage area weight is set based on the fruit tray spacing and the spiral guide rail speed, and the setting range is 0.5 ≤ ≤ 0.

7.

4. The device according to claim 2, wherein The classification model adopts YOLO v5 architecture; The local image is an RGB three-channel image with a resolution of 128×128 pixels; The detection results are defect category labels, including "no defect", "dent", "rot", and "worm-eaten"; The classification model is set to output detection results that meet a confidence threshold after pre-training, and the confidence threshold is 0.

8.

5. The device according to any one of claims 2 to 4, characterized in that The dynamic sorting control system is equipped with a redundant verification mechanism to verify the detection results of each node before triggering the sorting action; The redundant verification mechanism adopts a multi-node data cross-validation method, and calculates the comprehensive confidence of the detection results between multiple nodes through the following formula: ； in, is the comprehensive confidence of the detection results among multiple nodes, is the confidence of the detection result of the jth node, is the confidence weight of the jth node; in, ≥ 0.9 indicates a defect; The confidence of the detection results is dynamically adjusted according to the defect category label of the detection results. The confidence weight of "rot" is 1.2, the confidence weight of "worm-eaten" is 0.8, and the confidence weight of "sag" is 0.

9.

6. The device according to claim 2, characterized in that The parametric design of the spiral guide rail of the fruit tray meets the following conditions: The spacing between the spiral guide rails is between 20cm and 40cm, and the inclination angle of the spiral guide rails is between 0.4rad and 0.6rad; The friction coefficient between the top of the spiral guide rail and the contact surface of the fruit is dynamically adjusted according to the characteristics of the fruit skin. The friction coefficient of apples is 0.3 to 0.5, and the friction coefficient of citrus fruits is 0.1 to 0.

2.

7. The device according to claim 2, characterized in that The wide-angle lens coverage angle of each node in the multi-node top camera system is 180° to 220°, and the shooting areas of adjacent nodes are set to have an overlapping area of 10% to 30%.

8. The device according to claim 2, characterized in that The device supports modular expansion, and each channel is independently configured with a spiral fruit plate, a top camera, and an edge computing unit; The edge computing unit is deployed locally on the node to process images and transmit results in real time, reducing the load on the central server.

9. The device according to claim 2, characterized in that The device also includes a dynamic parameter adaptive adjustment module for automatically adjusting the spiral guide rail parameters and camera parameters according to the fruit type or surface characteristics; When the fruit type is apple, the dynamic parameter adaptive adjustment module sets the spacing of the spiral guide rail to 30 cm, the tilt angle to 0.5 rad, and the exposure time of the wide-angle lens to 1 / 1000 second; When the fruit type is orange, the dynamic parameter adaptive adjustment module sets the spacing of the spiral guide rail to 35 cm, the tilt angle to 0.45 rad, and the exposure time of the wide-angle lens to 1 / 500 second.

10. The device according to claim 9, characterized in that The dynamic parameter adaptive adjustment module calculates the fruit rotation angle using the following formula: ； in, is the fruit rotation angle, L is the length of the conveying path, r is the fruit radius, and α is the inclination angle of the spiral guide; The dynamic parameter adaptive adjustment module adjusts the spiral guide rail parameters and the camera parameters based on the standard of θ≥360°, so that the fruit to be inspected meets the inspection requirement of full surface coverage.

Images