A flexible mechanical palm heart coupling active spectrum in-situ detection device for apple quality

Abstract

The application discloses a flexible mechanical palm heart coupling active spectrum apple quality in-situ detection device, which comprises a flexible finger, an active multi-spectrum sensing system, a rudder module, a lower shell and a control box. The flexible finger is installed in cooperation with the positioning hole and the positioning pin on the shell. The multi-spectrum sensing system comprises a light-proof ring, a mechanical hand tray, a light uniformity sheet, a halogen lamp bead, a ring-shaped lamp plate, a light source bracket, a Fresnel lens, an AS7265X spectrum sensor and a mechanical hand base. The light uniformity sheet, the ring-shaped lamp plate, the Fresnel lens and the AS7265X are coaxially installed. The ring-shaped lamp plate is installed in the mechanical hand tray. The Fresnel lens and the AS7265X are installed below the light source bracket. The master control module is responsible for collecting, processing and analyzing fruit quality. A power supply module provides power supply for each module. The application overcomes the problem that it is difficult for the mechanical hand to obtain fruit quality in the picking process, and provides an effective technical means for on-site grading.

Description

Technical Field

[0001] This invention relates to the field of smart orchards, and in particular to an in-situ apple quality detection device using a flexible robotic hand palm-coupled active spectroscopy. Background Technology

[0002] In-situ acquisition of apple quality information is crucial for achieving precision agriculture and automated harvesting. Spectroscopic technology, with its real-time, rapid, and non-destructive advantages, shows great potential in the non-destructive detection of apple quality information (such as soluble solids content and firmness). Multispectral sensors, as a key implementation carrier of this technology, can directly detect the optical response of apples in specific wavelength bands and have become a powerful tool for on-site fruit quality analysis and grading in modern smart orchards. Existing research and practice show that, through reasonable wavelength selection and data analysis models, high quality prediction accuracy can be obtained under laboratory or controlled conditions, thus providing a basis for grading and traceability.

[0003] Patents CN109115708B, CN110263969B, and CN111220568B disclose an integrated non-destructive testing system and method for multiple internal qualities of apples, a dynamic prediction system and method for shelf-life apple quality, and an apple sugar content determination device and method based on near-infrared spectroscopy. The aforementioned detection devices or methods are all used for portable, offline collection and analysis of apple quality. However, these devices or methods are independent of automated harvesting processes and difficult to integrate. They cannot guarantee in-situ, online quality assessment of the fruit immediately after harvesting, easily leading to the need for secondary sorting after harvesting. This not only significantly reduces harvesting efficiency and increases operating costs, but also prevents real-time selective harvesting based on quality due to the separation of harvesting and testing, resulting in the mixed harvesting of high-quality fruit and economic losses. Therefore, there is an urgent need for a flexible palm-sized detection unit that integrates optics and mechanics, has active uniform illumination, short-range coaxial detection, and embedded real-time interpretation capabilities, in order to meet the requirements of in-situ, high-throughput, and repeatable quality assessment in automated harvesting processes. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of the aforementioned background technology and provide an in-situ apple quality detection device with active spectral coupling in the palm of a flexible robotic hand. This device coaxially integrates an active multispectral sensing system into the palm of the flexible robotic hand, enabling active illumination and multispectral information acquisition of apples at the same time and location when the fruit is grasped. Through a built-in radiation correction model and quality analysis algorithm, it interprets key quality indicators such as soluble solids content and hardness in real time, solving the problem that robotic hands cannot obtain fruit quality information in-situ and online during the harvesting process. This provides an efficient and integrated technical means for achieving quality-based real-time selective harvesting and on-site grading.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is: an in-situ apple quality detection device with flexible mechanical claw palm-coupled active spectrum, comprising a flexible finger, an active multispectral sensing system, a servo module, a lower shell, and a control box. The flexible finger is installed by engaging with positioning pins on the mechanical claw shell through positioning holes. The servo module is installed through a servo mounting bracket and is evenly installed in four positioning holes of the mechanical hand tray. The active multispectral sensing system is installed in the palm of the mechanical hand. The control box is installed below the active multispectral sensing system and includes a power supply module and a main control module. The lower shell is fixed to the base of the mechanical hand to support a stable connection between the active multispectral sensing system and the base of the mechanical hand.

[0006] Furthermore, the active multispectral sensing system includes a light-shielding ring, a robotic arm tray, a light-diffusing sheet, halogen lamp beads, a ring-shaped lamp panel, a light source bracket, a Fresnel lens, an AS7265X spectral sensor, and a robotic arm base. The light-diffusing sheet, ring-shaped lamp panel, Fresnel lens, and AS7265X multispectral sensor are coaxially mounted. The light-shielding ring is mounted on the upper side of the robotic arm tray. The ring-shaped lamp panel is mounted inside the robotic arm tray via the light source bracket. The Fresnel lens and AS7265X multispectral sensor are mounted below the light source bracket. The halogen lamp beads are arranged in a ring and welded to the ring-shaped lamp panel. The robotic arm tray, lower shell, and robotic arm base are sequentially fixedly connected from top to bottom along the axial direction by fasteners to form an integrated robotic arm shell.

[0007] Furthermore, the servo motor is mounted inside the servo motor housing via the servo motor mounting bracket. The servo motor drives the rotating shaft, thereby causing the servo disk to rotate. The servo disk drives the flexible fingers to move, so as to realize the grasping action of the robotic hand.

[0008] Furthermore, the AS7265X multispectral sensor includes three spectral sensors, AS72651, AS72652, and AS72653, which together cover 18 spectral bands: 410 nm, 435 nm, 460 nm, 485 nm, 510 nm, 535 nm, 560 nm, 585 nm, 610 nm, 645 nm, 680 nm, 705 nm, 730 nm, 760 nm, 810 nm, 860 nm, 900 nm, and 940 nm. The bandwidth is 20 nm, the average field of view of a single sensor is ±20.5°, and the three sensors are evenly distributed at a position 8.54 mm from the center point.

[0009] Furthermore, the halogen lamp beads have a light source half-value angle of 20°, a light source layout radius of 20mm, and a voltage of 3.3V. The eight halogen lamp beads are evenly arranged on the annular lamp plate at an inward tilt angle of 45°. The annular lamp plate is 30mm above the sample and is powered by a power supply module. The light-diffusing sheet, the annular lamp plate with eight halogen lamp beads welded on, the Fresnel lens, and the AS7265X multispectral sensor are installed sequentially from top to bottom along the axial direction. The positional relationship between each component is fixed and cannot be moved relative to each other.

[0010] Furthermore, the light-diffusing plate is made of quartz glass, with an outer diameter of 42 mm, an inner diameter of 16 mm, and a thickness of 1 mm. It is positioned and installed between the robotic arm tray and the light source bracket using a sleeve.

[0011] Furthermore, the Fresnel lens has a working surface angle of 32°, a diameter of 15mm, a focal length of 8.4mm, and is installed 20mm below the light homogenizer.

[0012] Furthermore, the power supply module provides power to the main control module and other modules; the main control module is connected to the AS7265X spectral sensor and the ring light panel via a wired connection, and is used to collect, process and interpret fruit quality information in situ, and transmit the quality information to the harvesting robot system to realize real-time on-site grading of the fruit.

[0013] Specifically, the operating steps of the device are as follows:

[0014] S1: Place a standard diffuse reflection calibration plate with 99% reflectance above the shading ring. Use the control terminal of the apple picking robot to click the automatic exposure function of the palm-mounted active multispectral sensing system and record the exposure time. Then, click the calibration plate acquisition function to save the calibration plate multispectral data to the main control module. Next, place a black calibration plate with 0% reflectance above the shading ring and click the dark background acquisition function to save the dark background multispectral data to the main control module. Finally, click the sample acquisition function to collect subsequent fruit multispectral data and perform radiometric correction on the original fruit multispectral data using the following formula:

[0015]

[0016] In the formula, R W Multispectral data for a 99% reflectance calibration plate. R D Multispectral data for a black calibration plate with 0% reflectance. I C This is the corrected multispectral data of the fruit.

[0017] S2: Based on the real-time corrected multispectral data of the fruit, and combined with the fruit quality analysis model in the main control module, real-time analysis of apple quality characteristics is achieved. The specific calculation formula is as follows:

[0018] The soluble solids content (SSC) of apples is calculated using the following formula:

[0019] Y SSC =13.67 - 7.38 × X1 - 21.49 × X2 + 39.02 × X3 - 57.33 × X4 + 16.91 × X5 + 35.92 × X6 - 56.49 × X7 + 42.96 × X8 - 56.11 × X9 + 51.73 × X 10

[0020] In the formula, Y SSC The parsed SSC values ​​of the fruit, X1 to X 10 These are the fruit reflectance values ​​obtained in step S1 after radiometric correction in the 410, 435, 460, 485, 560, 680, 705, 730, 760 and 940 nm bands, respectively.

[0021] Apple firmness is calculated using the following formula:

[0022] Y Firmness =10.91+130.15×X1-102.37×X2-115.88×X3-396.20×X4+1966.25×X5-2354.81×X6+971.17×X7

[0023] In the formula, Y Fimness X1 to X8 are the fruit hardness values ​​after analysis, respectively, and the radiometrically corrected fruit reflectance values ​​in the 435, 560, 730, 810, 860, 900 and 940 nm bands obtained in step S1.

[0024] S3: The parsing results of step S2 (including Y) SSC Y Fimness The received analysis results (and corresponding multispectral feature vectors) are transmitted to the main control system of the harvesting robot; the main control system generates and issues execution instructions based on the received analysis results to drive the execution end. The execution instructions are used to realize real-time classification and grasping decisions, specifically including but not limited to: when Y SSC With Y Fimness When the preset picking threshold is met, the system outputs a picking execution command and determines the grasping force, grasping posture, and grasping timing. When the preset threshold is not met, the system outputs an abandon or mark command for skipping or subsequent processing. When the analysis result has low confidence or is abnormal, a retest or manual review process is triggered. In addition, after picking is completed, the main control system stores the picking record (including picking time, location information, corresponding quality analysis data, and execution command summary) and can send it back to the background management system for subsequent data management and quality traceability.

[0025] The beneficial effects of this invention are as follows: By coaxially integrating an active multispectral sensing system into the palm of a flexible robotic hand, active illumination and multispectral information acquisition of apples are achieved at the same time and position when grasping the fruit. Through the built-in radiation correction model and quality analysis algorithm, key quality indicators such as soluble solids content and hardness are interpreted in real time. This solves the problem that robotic hands cannot obtain fruit quality information in situ and online during the picking process, and provides an efficient and integrated technical means for realizing quality-based real-time selective picking and on-site grading. Attached Figure Description

[0026] Figure 1 Schematic diagram of the overall structure of an in-situ apple quality detection device using active spectroscopy coupled to the palm of a flexible robotic arm.

[0027] Figure 2 Schematic diagram of an active multispectral sensing system

[0028] Figure 3 Cross-sectional view of an active multispectral sensing system

[0029] Figure 4 Schematic diagram of the servo module structure

[0030] Figure 5 Schematic diagram of the control system for an in-situ apple quality detection device using a flexible robotic hand palm-coupled active spectroscopy.

[0031] Figure 6 The reflection spectral curve after fruit radiation correction obtained for the device described in the invention example

[0032] Figure 7 The process of CARS selecting characteristic bands in the invention example ((a) Characteristic band selection for fruit SSC prediction, (b) Characteristic band selection for fruit hardness prediction)

[0033] Figure 8 The scatter plot of the apple quality analysis model constructed in the invention example ((a) SSC, (b) Hardness)

[0034] The markings in the figure are: 1 - Flexible finger, 2 - Active multispectral sensing system, 21 - Light shielding ring, 22 - Manipulator tray, 23 - Light homogenizing plate, 24 - Halogen lamp beads, 25 - Ring lamp board, 26 - Light source bracket, 27 - Fresnel lens, 28 - AS7265X spectral sensor, 29 - Manipulator base, 3 - Servo module, 31 - Servo mounting bracket, 32 - Rotating shaft, 33 - Servo disc, 34 - Servo servo, 35 - Servo housing, 4 - Lower shell, 5 - Control box, 51 - Power supply module, 52 - Main control module. Detailed implementation manners

[0035] The present invention will be described in detail below in conjunction with the accompanying drawings and specific embodiments.

[0036] As Figures 1 to 4 shown, this example provides an in-situ detection device for apple quality with flexible mechanical claw palm coupled with active spectrum, which includes a flexible finger 1, an active multispectral sensing system 2, a servo module 3, a lower shell 4, and a control box 5. The flexible finger 1 is installed by cooperating with the positioning pins on the manipulator shell through positioning holes. The servo module 3 is installed through a servo fixing bracket 31 and evenly installed in the four positioning holes of the manipulator tray 22. The active multispectral sensing system 2 is installed at the palm of the manipulator. The control box 5 is installed below the active multispectral sensing system 2. The control box 5 includes a power supply module 51 and a main control module 52, which are respectively responsible for supplying power to each sub-module and centralized control.

[0037] In the present invention, as Figure 2 and 3As shown, the active multi-spectral sensing system 2 includes an aperture stop 21, a manipulator tray 22, a light homogenizer 23, halogen lamp beads 24, an annular lamp board 25, a light source bracket 26, a Fresnel lens 27, an AS7265X spectral sensor 28, and a manipulator base 29. The light homogenizer 23, the annular lamp board 25, the Fresnel lens 27, and the AS7265X multi-spectral sensor 28 are coaxially installed. The annular lamp board 25 is installed inside the manipulator tray 22 through the light source bracket 26. The Fresnel lens 27 and the AS7265X multi-spectral sensor 28 are installed below the light source bracket 26, forming a coaxial light-emitting - lens - detection structure to ensure the light-receiving uniformity of the measured sample and the consistency of the detection field of view.

[0038] In the present invention, as Figure 4 shown, the servo steering gear 34 is installed inside the steering gear housing 35 through the steering gear fixing bracket 31. The servo steering gear 34 drives the rotating shaft 32, thereby driving the steering wheel 33 to rotate. The steering wheel 33 drives the flexible finger 1 to move, so as to realize the grasping of the fruit by the manipulator.

[0039] In the present invention, the AS7265X multi-spectral sensor 28 includes three spectral sensors, namely AS72651, AS72652, and AS72653, which together cover 18 spectral bands of 410 nm, 435 nm, 460 nm, 485 nm, 510 nm, 535 nm, 560 nm, 585 nm, 610 nm, 645 nm, 680 nm, 705 nm, 730 nm, 760 nm, 810 nm, 860 nm, and 940 nm. The bandwidth is 20 nm. The average field of view angle of a single sensor is ±20.5°, and the three sensors are evenly distributed at a position 8.54 mm away from the center point.

[0040] In the present invention, the light source half-value angle of the halogen lamp bead 24 is 20°, the light source layout radius is 20 mm, the voltage is 3.3 V. The 8 halogen lamp beads are evenly arranged on the annular lamp board 25 at an angle of 45° inward inclination. The annular lamp board 25 is 30 mm away from the sample height and is powered by the power supply module 51, providing stable active illumination conditions for the multi-spectral sensor.

[0041] In the present invention, as Figure 2 and 3 shown, the light homogenizer 23 is made of quartz glass, with an outer diameter of 42 mm, an inner diameter of 16 mm, and a thickness of 1 mm. It is installed between the manipulator tray 22 and the light source bracket 26 through sleeve positioning to ensure the coaxial positioning accuracy with the light source, thereby improving the spatial uniformity of active illumination and providing mechanical protection and light shielding for the internal optical and electronic modules.

[0042] In the present invention, as Figure 2 and 3 shown, the working surface angle of the Fresnel lens 27 is 32°, the diameter is 15 mm, the focal length is 8.4 mm, and it is installed 20 mm below the light homogenizing plate 23; this lens is used to collimate and focus the diffuse reflection light on the fruit surface to improve the consistency of the sensor reception and enhance the light collection efficiency.

[0043] In the present invention, the power supply module 51 provides power for the main control module 52 and other modules; the main control module 52 is connected to the AS7265X spectral sensor 28 and the annular lamp board 25 in a wired manner, responsible for the acquisition, preprocessing of multi-spectral data and in-situ fruit quality analysis based on the built-in model, and transmits the analysis results to the main control system of the picking robot for real-time grading and grasping decision-making.

[0044] Figure 5 The block diagram of the present device is given, including a host computer, a main control module, a power supply module, a servo module, a Bluetooth module, and an active multi-spectral sensing system. Among them, the active multi-spectral sensing system and the servo module are both connected to the main control module to transmit data to it, and the power supply module is connected to the main control module to supply power to it. The active multi-spectral sensing system includes a light source module, a condenser module, and a spectral detector. In this embodiment, the main control module transmits data to the host computer through the Bluetooth module.

[0045] The operation steps of the said device are as follows:

[0046] S1: Place a standard diffuse reflection calibration plate with a reflectivity of 99% above the light shielding ring, click the automatic exposure function of the palm active multi-spectral sensing system of the apple picking robot, and record the exposure time; then click the calibration plate acquisition function, and save the multi-spectral data of the calibration plate to the main control module; then, place a black calibration plate with a reflectivity of 0% above the light shielding ring, click the dark background acquisition function, and save the multi-spectral data of the dark background to the main control module; finally, click the sample acquisition function to collect the subsequent multi-spectral data of the fruit, and perform radiation correction on the original multi-spectral data of the fruit through the following formula:

[0047]

[0048] In the formula, R W is the multi-spectral data of the 99% reflectivity calibration plate, R D is the multi-spectral data of the 0% reflectivity black calibration plate, I C is the corrected multi-spectral data of the fruit.

[0049] S2: Based on the real-time corrected fruit multispectral data, combined with the fruit quality analysis model in the main control module, realize the real-time analysis of apple quality characteristics.

[0050] S3: Transmit the analysis results of step S2 (including Y SSC , Y Fimness and the corresponding multispectral feature vectors) to the main control system of the picking robot; the main control system generates and issues execution instructions according to the received analysis results to drive the execution end, and the execution instructions are used to achieve real-time grading and grasping decisions, specifically including but not limited to: when Y SSC and Y Fimness meet the preset picking threshold, output the picking execution instruction and determine the grasping force, grasping posture and grasping timing; when not meeting the preset threshold, output the abandonment or marking instruction to skip or perform subsequent processing; when the analysis result has low confidence or is abnormal, trigger the retest or manual review process. In addition, after picking, the main control system stores the picking records (including picking time, location information, corresponding quality analysis data and execution instruction summary) and can be transmitted back to the background management system for subsequent data management and quality traceability.

[0051] In step S1, 100 apples (variety: Fuji) were selected as the research objects in the experimental base of Shandong Agricultural University, and the device was automatically exposed using a standard diffuse reflection calibration plate with a reflectivity of 99%. The exposure time and gain were 280 ms and 2 dB respectively; then the multispectral data of the standard diffuse reflection calibration plate were collected and recorded, and the dark background data were obtained by covering the lens cap. Finally, the original data of the 100 apples collected were radiometrically corrected according to the following formula to obtain the reflectivity of the fruit, as Figure 6 shown.

[0052]

[0053] In the formula, R W is the multispectral data of the 99% reflectivity calibration plate, R D is the multispectral data of the 0% reflectivity black calibration plate, I C is the multispectral data of the corrected fruit.

[0054] In step S2, after the spectral data acquisition, the flesh firmness was measured using a digital fruit hardness meter (GY-4, Sanliang Technology Co., ltd, Guangzhou, China) equipped with a stainless steel probe with a diameter of 3.2 mm. The probe penetrated vertically into the flesh at an acceleration of 5 mm / s until a penetration depth of 10 mm was reached, and the hardness value displayed by the instrument at this time was recorded. The average value of the hardness values at four equally spaced measurement points was used as the reference hardness value. After the hardness measurement, flesh samples were taken from the four measurement positions, and juice was extracted using a juicer. The juice was dropped onto the prism plate of a digital refractometer (PAL-1, ATAGO, Japan) using a dropper, and the SSC was measured. The average value of the four measurement values was used as the SSC reference value.

[0055] In step S2, based on the sample set partitioning (SPXY) algorithm for joint X–Y distance, all samples were divided into a modeling set and a prediction set at a ratio of 7:3. The SPXY algorithm considered the distribution characteristics of the independent variable spectral values and the dependent variable FCQI values simultaneously during the partitioning process, which could ensure the consistency of the modeling set and the prediction set in the spectral and target value spaces, thereby improving the generalization ability and prediction stability of the model. Subsequently, the competitive adaptive reweighted sampling (CARS) method was used to optimize the characteristic bands for predicting the fruit SSC and firmness. CARS constructs sampling weights based on the absolute values of the partial least squares (PLS) regression coefficients, and uses an iterative weighted sampling strategy to dynamically adjust the selection probabilities of each band, so that the bands that contribute the most to the prediction performance are preferentially retained. Figure 7 The process of selecting characteristic bands using the CARS algorithm is shown. As the number of sampling runs increases, the number of bands shows a decreasing trend. For SSC and firmness predictions, 10 and 7 bands were selected respectively, and the specific bands are shown in Table 1.

[0056] Table 1 Characteristic bands optimized using the CARS algorithm for predicting fruit SSC and firmness

[0057] index Number band / nm SSC 10 410，435，460，485，560，680，705，730，760，940 hardness 7 435，560，730，810，860，900，940

[0058] In step S2, taking the spectral reflectance of the characteristic bands as the input, combined with the SSC and firmness values of the fruit respectively, a multivariate linear regression algorithm was used to construct an analytical model. As Figure 8 shown, for the fruit SSC, the determination coefficient and root mean square error RMSEC of the modeling set were 0.86 and 0.4419 respectively, and the determination coefficient and root mean square error RMSEP of the prediction set were 0.86 and 0.3984 respectively. For the fruit firmness, the RMSEC of the modeling set was 0.83 and 1.0960 respectively, and the RMSEP of the prediction set was 0.80 and 0.9147 respectively. The scatter plot of the analytical model is shown in Figure 8 shown.

[0059] In step S2, the constructed analytical model for the fruit SSC is as follows:

[0060] Y SSC = 13.67 - 7.38×X1 - 21.49×X2 + 39.02×X3 - 57.33×X4 + 16.91×X5 + 35.92×X6 - 56.49×X7 + 42.96×X8 - 56.11×X9 + 51.73×X 10

[0061] In the formula, Y SSC is the value of the fruit SSC after analysis, and X1 to X 10 are respectively the fruit reflectance values at 410, 435, 460, 485, 560, 680, 705, 730, 760, and 940 nm bands after radiation correction obtained in step S1.

[0062] The apple hardness is calculated by the following formula:

[0063] Y Firmness = 10.91 + 130.15×X1 - 102.37×X2 - 115.88×X3 - 396.20×X4 + 1966.25×X5 - 2354.81×X6 + 971.17×X7

[0064] In the formula, Y Fimness is the value of the fruit hardness after analysis, and X1 to X8 are respectively the fruit reflectance values at 435, 560, 730, 810, 860, 900, and 940 nm bands after radiation correction obtained in step S1.

[0065] The above results show that the analytical model constructed in this embodiment can accurately obtain the soluble solid content and hardness information of apples, providing a technical basis for in-situ fruit perception and on-site grading.

[0066] The scope of protection of the claims shall be subject to the appended claims. For those skilled in the art, without departing from the spirit of the present invention and the limitations of the claims, the structural dimensions, sensor models, light source parameters, optical component configurations, analytical models, etc. of the above embodiments can be appropriately adjusted or replaced. For example, replacing them with spectral chips with equivalent bandwidth and band coverage, using homogenizing films of different materials, or adjusting the light source layout, etc. These equivalent or improved modifications all fall within the protection scope of the present invention.

Claims

1. A flexible mechanical claw palm-coupled active spectroscopy in-situ detection device for apple quality, characterized in that... It includes a flexible finger (1), an active multispectral sensing system (2), a servo module (3), a lower shell (4), and a control box (5). The flexible finger (1) is installed by cooperating with the positioning pin on the mechanical claw shell through the positioning hole. The servo module (3) is installed by the servo mounting bracket (31) and is evenly installed in the four positioning holes of the robotic arm tray (22). The active multispectral sensing system (2) is installed in the palm of the robotic arm. The control box (5) is installed below the active multispectral sensing system (2). The control box (5) includes a power supply module (51) and a main control module (52). The lower shell (4) is fixed to the robotic arm base (29) to support the stable connection between the active multispectral sensing system and the robotic arm base (29). The active multispectral sensing system (2) includes a light shield (21), a robotic arm tray (22), a light diffuser (23), halogen lamp beads (24), a ring lamp panel (25), a light source bracket (26), a Fresnel lens (27), an AS7265X spectral sensor (28), and a robotic arm base (29). The light diffuser (23), the ring lamp panel (25), the Fresnel lens (27), and the AS7265X multispectral sensor (28) are coaxially mounted. The light shield (21) is mounted on the robotic arm tray. (22) On the upper side, the annular lamp plate (25) is installed inside the robotic arm tray (22) through the light source bracket (26). The Fresnel lens (27) and the AS7265X multispectral sensor (28) are installed below the light source bracket (26). The halogen lamp beads (24) are arranged in a ring and welded on the annular lamp plate (25). The robotic arm tray (22), the lower shell (4) and the robotic arm base (29) are fixedly connected from top to bottom along the axial direction by fasteners to form an integrated robotic arm shell. The servo module (3) includes a servo mounting bracket (31), a drive rotation shaft (32), a servo disk (33), a servo servo (34), and a servo housing (35). The servo servo (34) is installed in the servo housing (35) through the servo mounting bracket (31). The servo servo (34) drives the rotation shaft (32), thereby driving the servo disk (33) to rotate. The servo disk (33) drives the flexible finger (1) to move, so as to realize the gripping action of the robotic hand. The halogen lamp beads (24) have a light source half-value angle of 20°, a light source layout radius of 20mm, and a voltage of 3.3V. Eight halogen lamp beads (24) are evenly arranged on the ring lamp plate (25) at an inward tilt angle of 45°. The ring lamp plate (25) is 30mm above the sample and is powered by a power supply module (51). The light homogenizer (23), the ring lamp plate (25) with eight halogen lamp beads (24) welded on, the Fresnel lens (27), and the AS7265X multispectral sensor (28) are installed sequentially from top to bottom along the axial direction. The positional relationship between each component is fixed and cannot be moved relative to each other.

2. The apparatus according to claim 1, characterized in that... The AS7265X multispectral sensor (28) includes three spectral sensors, AS72651, AS72652 and AS72653, which together cover 18 spectral bands: 410 nm, 435 nm, 460 nm, 485 nm, 510 nm, 535 nm, 560 nm, 585 nm, 610 nm, 645 nm, 680 nm, 705 nm, 730 nm, 760 nm, 810 nm, 860 nm, 900 nm and 940 nm. The bandwidth is 20 nm, the average field of view of a single sensor is ±20.5°, and the three sensors are evenly distributed at a position 8.54 mm away from the center point.

3. The apparatus according to claim 1, characterized in that... The light-diffusing plate (23) is made of quartz glass, with an outer diameter of 42 mm, an inner diameter of 16 mm, and a thickness of 1 mm. It is positioned and installed between the robotic arm tray (22) and the light source bracket (26) by a sleeve.

4. The apparatus according to claim 1, characterized in that... The Fresnel lens (27) has a working surface angle of 32°, a diameter of 15 mm, a focal length of 8.4 mm, and is installed 20 mm below the light homogenizer (23).

5. The apparatus according to claim 1, characterized in that... The power supply module (51) provides power to the main control module (52) and other modules. The main control module (52) is connected to the AS7265X spectral sensor (28) and the ring light board (25) via wired connection. It is used to collect, process and interpret fruit quality information in situ, and transmit the quality information to the picking robot system to realize real-time on-site grading of the fruit.

6. The apparatus according to any one of claims 1-5, characterized in that... The operating steps of the device are as follows: S1: Place a standard diffuse reflection calibration plate with a reflectivity of 99% above the light-shielding ring, and use the automatic exposure function of the palm active multispectral sensing system on the control end of the apple picking robot to record the exposure time; Next, click the calibration plate acquisition function to save the calibration plate multispectral data to the main control module; then, place the black calibration plate with 0% reflectance above the light-shielding ring, click the dark background acquisition function, and save the dark background multispectral data to the main control module; finally, click the sample acquisition function to acquire subsequent fruit multispectral data, and use the following formula to obtain the raw fruit multispectral data. Radiation correction: In the formula, R W For multispectral data of a 99% reflectance calibration plate, R D For multispectral data of a black calibration plate with 0% reflectivity, I C The corrected multispectral data of the fruit; S2: Based on the real-time corrected multispectral data of the fruit, and combined with the fruit quality analysis model in the main control module, real-time analysis of apple quality characteristics is achieved. The specific calculation formula is as follows: The soluble solids content (SSC) of apples is calculated using the following formula: AND SSC =13.67-7.38×X1-21.49×X2+39.02×X3-57.33×X4+16.91×X5+35.92×X6-56.49×X7+42.96×X8-56.11×X9+51.73×X 10 In the formula, Y SSC The parsed SSC values ​​of the fruit, X1 to X 10 These are the fruit reflectance values ​​obtained in step S1 after radiometric correction in the 410, 435, 460, 485, 560, 680, 705, 730, 760 and 940 nm bands, respectively. Apple firmness is calculated using the following formula: AND Firmness =10.91+130.15×X1-102.37×X2-115.88×X3-396.20×X4+1966.25×X5-2354.81×X6+971.17×X7 In the formula, Y Fimness The fruit hardness values ​​are the analyzed values, and X1 to X8 are the fruit reflectance values ​​obtained in step S1 after radiometric correction in the 435, 560, 730, 810, 860, 900 and 940 nm bands, respectively. S3: The parsing result of step S2 is transmitted to the main control system of the picking robot; the main control system generates and issues execution instructions based on the received parsing result to drive the execution end, and the execution instructions are used to realize real-time grasping decision.

7. The apparatus according to claim 6, characterized in that... The specific decision-making process includes: when Y SSC With Y Fimness When the preset picking threshold is met, a picking execution command is output and the grasping force, grasping posture and grasping timing are determined; when the preset threshold is not met, an abandon or marking command is output for skipping or subsequent processing; when the analysis result has low confidence or is abnormal, a retest or manual review process is triggered.

Images