Multi-vision fused multi-stage discrete corn breakage rate accurate detection device and breakage rate calculation method

Abstract

The invention discloses a multi-vision fusion multi-stage discrete corn breakage rate accurate detection device and a breakage rate calculation method. The device comprises a shell, a multi-stage discrete mechanism and a multi-vision fusion system, the multi-stage dispersing mechanism comprises a corn grain discharging device, a feeding inclined plate, a material blocking inclined plate, a vibrating plate and a conveying belt, the corn grain discharging device is used for conveying corn samples to the feeding inclined plate in a single-row and single-layer mode, the material blocking inclined plate is arranged above the material blocking inclined plate and used for preliminarily dispersing the corn samples, and the vibrating plate is arranged below the material blocking inclined plate and used for secondarily dispersing the corn samples; the conveying belt is arranged below the vibrating plate; the multi-view vision fusion system comprises a pre-detection camera, an upper detection camera and a lower detection camera, multi-angle and all-directional detection of the breakage rate of the corn kernel groups can be achieved, the detection precision and the coverage rate are improved, and blind areas are effectively reduced. Meanwhile, the U-Net segmentation and dynamic data fusion algorithm is adopted, the crushing rate calculation precision is improved, and the method has the advantages of being efficient, accurate, high in anti-interference performance and the like.

Description

Technical Field

[0001] The present invention relates to the technical field of agricultural machinery, and in particular to a multi-stage discrete corn breakage rate precision detection device and a breakage rate calculation method based on multi-vision fusion. Background Art

[0002] Corn breakage rate is a key indicator of mechanized corn harvesting quality. Kernels with high breakage rates are susceptible to mold, directly impacting harvest yield, quality, and storage safety. Traditional detection methods rely primarily on manual sampling and laboratory analysis, which are subject to inefficiency, time-consuming, high-cost, high subjectivity, and poor real-time performance. Therefore, to improve the accuracy of quantitative kernel breakage rate detection, research and improvement of corn breakage rate detection devices are crucial.

[0003] With the development of computer vision and image processing technologies, automated vision-based detection methods have become a research hotspot. Existing corn breakage rate detection devices use a single vision system, which is susceptible to interference from factors such as lighting and occlusion in complex environments, making high-precision detection difficult. Furthermore, corn samples in traditional detection devices tend to stack during transport, making it difficult to achieve a single layer, which affects detection accuracy. Summary of the Invention

[0004] Purpose of the invention: The purpose of the present invention is to provide a multi-stage discrete corn breakage rate precision detection device and a breakage rate calculation method with simple structure and reasonable design and multi-eye vision fusion, which can improve the accuracy of corn breakage rate detection and overcome the stacking problem during corn sample transmission.

[0005] Technical solution: A multi-stage discrete corn breakage rate precision detection device with multi-vision fusion of the present invention includes a shell, a multi-stage discrete mechanism and a multi-vision fusion system; the multi-stage discrete mechanism includes a corn pelletizer, a feed inclined plate, a baffle inclined plate, a vibration plate and a transparent conveyor belt, the corn pelletizer is arranged on the top of the shell to transport the corn sample in a single row and single layer to the feed inclined plate, the feed inclined plate and the baffle inclined plate are inclined and relatively arranged on both sides of the shell, the feed inclined plate is arranged above the baffle inclined plate for preliminarily discretizing the corn sample, the vibration plate is arranged below the baffle inclined plate for secondary discretization of the corn sample, and the transparent conveyor belt is arranged below the vibration plate; the multi-vision fusion system includes an image processing system, a pre-detection camera arranged on the top of the shell, an upper detection camera arranged on the shell and a lower detection camera arranged at the bottom of the shell.

[0006] Furthermore, the pre-detection camera is located above the front end of the transparent conveyor belt, the upper detection camera is located above the rear end of the transparent conveyor belt, and the lower detection camera is located below the middle of the transparent conveyor belt.

[0007] Furthermore, the feed inclined plate and / or the baffle inclined plate are movably connected to the shell, and the upper ends of the feed inclined plate and / or the baffle inclined plate are provided with an electric telescopic rod, which is used to push the feed inclined plate and / or the baffle inclined plate to swing.

[0008] Furthermore, the pre-detection camera, upper detection camera and lower detection camera are respectively connected to the shell through camera fixing plates, and the pre-detection camera, upper detection camera and lower detection camera are all equipped with an annular LED light source and installed in the inner diameter circle of the annular LED light source.

[0009] Furthermore, a vibration motor is provided at the bottom of the vibration plate, and the vibration plate is connected to the housing via a spring.

[0010] Furthermore, side protection shells are installed on the outer sides of the material blocking inclined plate and the vibration plate.

[0011] A multi-level discrete corn breakage rate calculation method based on multi-view fusion of the present invention comprises the following steps:

[0012] S1: Acquire multi-view images of corn samples through a multi-view vision system;

[0013] S2: The U-Net semantic segmentation algorithm is used to distinguish between intact corn and broken corn areas, and the DiceLoss loss function is used to optimize the segmentation accuracy of overlapping areas;

[0014] S3: Dynamically fuse the detection results of multiple cameras and select the final fragmentation rate value based on the error threshold.

[0015] Furthermore, the U-Net semantic segmentation algorithm is used in S2 to distinguish between intact corn and broken corn areas, and the DiceLoss loss function is used to optimize the segmentation accuracy of overlapping areas. The formula is: Among them, y i and They represent the label value and predicted value of pixel i respectively, and N is the total number of pixels, which is equal to the number of pixels in a single image multiplied by batchsize.

[0016] Furthermore, in S1, multi-view corn images are synchronously acquired by the pre-detection camera, the upper detection camera, and the lower detection camera; in S3, when the error between the corn breakage rate result obtained by the pre-detection camera and the average value of the corn breakage rates obtained by the upper and lower detection cameras is less than or equal to 5%, the corn breakage rate result obtained by the pre-detection camera is ignored, and the average value of the corn breakage rates obtained by the upper and lower detection cameras is taken as the final result; when the error between the corn breakage rate result obtained by the pre-detection camera and the average value of the corn breakage rates obtained by the upper and lower detection cameras is greater than 5%, the average value of the corn breakage rates of the pre-detection camera, the upper detection camera, and the lower detection camera is taken as the final result.

[0017] Beneficial effects: Compared with the prior art, the present invention has the following advantages:

[0018] (1) The present invention adopts multi-view vision to obtain more comprehensive information through multiple perspectives, thereby detecting the breakage rate of the obscured back of corn, reducing the blind spots and errors of a single perspective, and at the same time, integrating the breakage rate results of multiple perspective images to help improve the stability and accuracy of the results of the entire detection process.

[0019] (2) The present invention adopts a multi-stage discrete structure, and a corn granulator is installed at the entrance. After the corn kernel samples enter the corn granulator, they are rotated into the detection device in rows through the grooves inside the corn granulator, which is conducive to the single-row and single-layer transportation of the corn detection samples. After being sent into the detection device, the electric telescopic rod is used in conjunction with the feed ramp and the material blocking ramp for preliminary discreteness, and the subsequent vibration motor and the vibration plate are used for further discreteness, effectively making the corn samples single-layered, reducing the stacking and adhesion of the corn samples, and improving the detection accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 This is a schematic diagram of the overall external structure of the present invention;

[0021] Figure 2 Schematic diagram of the internal structure of the present invention;

[0022] Figure 3 This is a schematic diagram of the multi-level discrete structure of the present invention;

[0023] Figure 4 This is a schematic diagram of the multi-eye vision structure of the present invention;

[0024] 1-Corn grain discharger, 2-Top cover, 3-Shell, 4-Pre-detection camera, 5-Electric telescopic rod, 6-Feeding ramp, 7-Blocking ramp, 8-Vibration plate, 9-Spring, 10-Vibration motor, 11-Power supply, camera fixing plate, 12-LED light source, 13-Upper detection camera, 14-Conveyor belt, 15-Base, 16-Lower detection camera, 17-Side protective shell. DETAILED DESCRIPTION

[0025] The technical solution of the present invention will be further described below with reference to the accompanying drawings.

[0026] The invention discloses a multi-stage discrete corn breakage rate precision detection device with multi-vision fusion. Figure 1-4 The shell assembly, multi-level discrete mechanism and multi-viewing fusion system shown.

[0027] The housing assembly includes the housing 3, top cover 2, and base 15. The top cover 2 and base 15 are fixed to the housing 3 using bolts and nuts. Side protective covers 17 are screwed to the sides of the housing 3. The main body of the housing 3 and the base 15 are firmly connected, forming the main frame of the device, providing a mounting base and protection for the internal components.

[0028] The multi-stage discretization mechanism includes a corn displacer 1, an infeed ramp 6, a baffle ramp 7, a vibrating plate 8, and a transparent conveyor belt 14. The corn displacer 1 is screwed onto the top cover 2 and the housing 3. It has internal grooves. After corn kernel samples enter the corn displacer 1, they are rotated through the grooves and fed row by row into the interior of the testing device, namely onto the infeed ramp 6, achieving single-file, single-layer conveyance of the corn samples. The infeed ramp 6 and baffle ramp 7 are arranged at an angle relative to each other on opposite sides of the housing 3. The infeed ramp 6 is positioned above the baffle ramp 7 and is used for initial discretization of the corn samples. The upper ends of the infeed ramp 6 and baffle ramp 7 are equipped with electric telescopic rods 5. These rods are mounted above the infeed ramp 6 and baffle ramp 7, respectively, and propel the infeed ramp 6 and baffle ramp 7 back and forth horizontally between 30° and 45°. This allows the infeed ramp 6 and baffle ramp 7 to vibrate the corn samples while draining the corn, effectively discretizing the corn samples. The vibration plate 8 is mounted below the feed ramp 6 and the baffle ramp 7, and above the conveyor belt 14. The vibration plate 8 is connected to the housing 3 via a spring 9, and a vibration motor 10 is provided at the bottom. When the corn sample passes through the feed ramp 6 and the baffle ramp 7 and undergoes initial vibration and dispersion, it falls onto the vibration plate 8. The vibration motor 10 drives the vibration plate 8 and cooperates with the spring 9 to achieve vibration, further dispersing the corn sample and achieving more efficient single-layering of the sample. A transparent conveyor belt 14 is provided below the vibration plate 8 to transport the dispersed corn sample to the testing area. Its transparency facilitates the lower detection camera 16 to inspect the back of the corn sample through it.

[0029] The multi-eye vision fusion system includes a pre-detection camera 4, an upper detection camera 13 and a lower detection camera 16. The pre-detection camera 4 is installed below the top cover 2, above the feed ramp 6 and the baffle ramp 7, and at the front section of the transparent conveyor belt 14 to perform a preliminary pre-detection of the breakage rate of the corn detection sample. The lower detection camera 16 is installed above the base 15, below the transparent conveyor belt 14, and in the middle section of the transparent conveyor belt 14 to further detect the breakage rate of the back of the corn sample through the transparent conveyor belt 14. The upper detection camera 13 is installed below the power supply and camera fixing plate 11, above the transparent conveyor belt 14, and at the rear section of the transparent conveyor belt 14 to perform the final step of detection of the breakage rate of the corn sample on the detection conveyor belt 14. The three detection cameras are all connected to the housing 3 through the camera fixing plate, and are all equipped with an annular LED light source and installed within the inner diameter ring of the annular LED light source to reduce the impact of lighting conditions and complex environments on the image.

[0030] The corn sample to be tested is poured into the corn pelletizer 1. The corn kernel samples are rotated row by row through the grooves inside the corn pelletizer 1 and fed into the interior of the detection device, that is, onto the feed ramp 6, to achieve single-row and single-layer transportation of the corn test samples. Corn that does not need to be tested can flow into the granary from the side discharge port of the corn pelletizer 1. When the corn test sample passes through the feed ramp 6 and the baffle ramp 7, the electric telescopic rod 5 stretches back and forth, causing the feed ramp 6 and the baffle ramp 7 to swing back and forth between 30° and 45° in the horizontal direction, draining and vibrating the corn sample to achieve preliminary discretization. After preliminary discretization, the corn sample falls onto the vibration plate 8. The vibration motor 10 drives the vibration plate 8 and cooperates with the spring 9 to achieve vibration, further discretizing the corn test sample and improving the efficiency of sample single layering.

[0031] Pre-detection camera 4 performs a preliminary inspection of the corn sample's breakage rate at the front section of transparent conveyor belt 14. Lower detection camera 16, located in the middle section of transparent conveyor belt 14, inspects the breakage rate from the back of the corn sample through the transparent conveyor belt 14. Upper detection camera 13, located at the rear section of transparent conveyor belt 14, performs a final inspection of the corn sample's breakage rate. When capturing images, the three detection cameras' accompanying ring-shaped LED light sources operate simultaneously, minimizing the impact of lighting conditions and complex environments on the images.

[0032] The collected multi-view corn images are transmitted to the image processing system. For the acquired images, a deep learning model is subsequently used to segment the intact corn and broken corn using the U-Net semantic segmentation algorithm. The proportion of the broken corn pixel area in the entire corn test sample pixel area is calculated to obtain the corn breakage rate result. For the overlapping areas of the segmentation, the DiceLoss loss function is used for optimization, that is, Among them, y i and They represent the label value and predicted value of pixel i respectively, and N is the total number of pixels, which is equal to the number of pixels in a single image multiplied by batchsize.

[0033] After being processed by the deep learning image algorithm, the detection results of the three detection cameras are dynamically fused. For the data obtained by multiple detection cameras, in order to obtain a more accurate corn breakage rate, it is stipulated that when the error between the corn breakage rate result obtained by the pre-detection camera and the average corn breakage rate obtained by the upper and lower detection cameras is less than or equal to 5%, the corn breakage rate result obtained by the pre-detection camera is ignored, and the average corn breakage rate obtained by the upper and lower detection cameras is taken as the final result; when the error between the corn breakage rate result obtained by the pre-detection camera and the average corn breakage rate obtained by the upper and lower detection cameras is greater than 5%, the average corn breakage rate of the pre-detection camera, upper detection camera, and lower detection camera is taken as the final result.

Claims

1. A multi-stage discrete corn breakage rate precision detection device with multi-vision fusion, comprising a housing (3), characterized in that: The invention also includes a multi-stage discrete mechanism and a multi-view visual fusion system; the multi-stage discrete mechanism includes a corn granulator (1), a feed inclined plate (6), a baffle inclined plate (7), a vibration plate (8) and a transparent conveyor belt (14); the corn granulator (1) is arranged on the top of the shell (3) for conveying the corn sample to the feed inclined plate (6) in a single row and single layer; the feed inclined plate (6) and the baffle inclined plate (7) are arranged on both sides of the shell (3) at an angle relative to each other; the feed inclined plate (6) is arranged above the baffle inclined plate (7) for preliminarily discrete corn samples; the vibration plate (8) is arranged below the baffle inclined plate (7) for secondary discrete corn samples; and the transparent conveyor belt (14) is arranged below the vibration plate (8); the multi-view visual fusion system includes an image processing system, a pre-detection camera (4) arranged on the top of the shell (3), an upper detection camera (13) arranged on the shell (3) and a lower detection camera (16) arranged on the bottom of the shell (3).

2. The multi-level discrete corn breakage rate precision detection device based on multi-vision fusion according to claim 1 is characterized in that: The pre-detection camera (4) is located above the front end of the transparent conveyor belt (14), the upper detection camera (13) is located above the rear end of the transparent conveyor belt (14), and the lower detection camera (16) is located below the middle of the transparent conveyor belt (14).

3. The multi-level discrete corn breakage rate precision detection device based on multi-vision fusion according to claim 1 is characterized in that: The feed inclined plate (6) and / or the baffle inclined plate (7) are movably connected to the housing, and an electric telescopic rod (5) is provided at the upper end of the feed inclined plate (6) and / or the baffle inclined plate (7), and the electric telescopic rod (5) is used to push the feed inclined plate (6) and / or the baffle inclined plate (7) to swing.

4. The multi-stage discrete corn breakage rate precision detection device based on multi-vision fusion according to claim 1 or 2, characterized in that: The pre-detection camera (4), the upper detection camera (13) and the lower detection camera (16) are respectively connected to the housing (3) via a camera fixing plate (11); the pre-detection camera (4), the upper detection camera (13) and the lower detection camera (16) are all equipped with an annular LED light source (12) and are installed within the inner diameter circle of the annular LED light source (12).

5. The multi-level discrete corn breakage rate precision detection device based on multi-vision fusion according to claim 1 is characterized in that: A vibration motor (10) is provided at the bottom of the vibration plate (8), and the vibration plate (8) is connected to the housing (3) via a spring (9).

6. The multi-level discrete corn breakage rate precision detection device based on multi-vision fusion according to claim 1 is characterized in that: The outer sides of the material blocking inclined plate (7) and the vibration plate (8) are also provided with side protection shells (17).

7. A multi-level discrete corn breakage rate calculation method based on multi-view fusion, characterized in that: The following steps are involved: S1: Acquire multi-view images of corn samples through a multi-view vision system; S2: The U-Net semantic segmentation algorithm is used to distinguish between intact corn and broken corn areas, and the DiceLoss loss function is used to optimize the segmentation accuracy of overlapping areas; S3: Dynamically fuse the detection results of multiple cameras and select the final fragmentation rate value based on the error threshold.

8. The multi-level discrete corn breakage rate calculation method based on multi-view fusion according to claim 7 is characterized in that: In S2, the U-Net semantic segmentation algorithm is used to distinguish between intact corn and broken corn areas, and the DiceLoss loss function is used to optimize the segmentation accuracy of overlapping areas. The formula is: Among them, y i and They represent the label value and predicted value of pixel i respectively, and N is the total number of pixels, which is equal to the number of pixels in a single image multiplied by batchsize.

9. The multi-level discrete corn breakage rate calculation method based on multi-view fusion according to claim 7 is characterized in that: In S1, multi-view corn images are synchronously acquired by the pre-detection camera, the upper detection camera, and the lower detection camera; in S3, when the error between the corn breakage rate result obtained by the pre-detection camera and the average value of the corn breakage rates obtained by the upper and lower detection cameras is less than or equal to 5%, the corn breakage rate result obtained by the pre-detection camera is ignored, and the average value of the corn breakage rates obtained by the upper and lower detection cameras is taken as the final result; when the error between the corn breakage rate result obtained by the pre-detection camera and the average value of the corn breakage rates obtained by the upper and lower detection cameras is greater than 5%, the average value of the corn breakage rates of the pre-detection camera, the upper detection camera, and the lower detection camera is taken as the final result.

Images