Method and device for detecting seed metering performance of seed metering device based on high frame line scanning

Abstract

The invention relates to a seed-metering device performance detection method and device based on high-frame line scanning, and is suitable for performance detection of a seed-metering device. According to the detection method, the linear array camera is used for carrying out real-time image information acquisition on the falling state of the seed flow in the seed metering process, and then an image is transmitted to the image processing system for information processing and analysis; the information processing and analysis comprises the steps of accurately obtaining a time sequence space distance of a seed-metering seed flow through image preprocessing and training based on an improved deep learning network model, analyzing and calculating to obtain key performance indexes of the seed-metering device, such as a qualified index, a miss-seeding index and a reseeding index, and realizing seed-metering performance detection of the seed-metering device; the device applied by the invention comprises a rack, a seed metering device, an image acquisition device, an image processing device, a seed receiving disc and an integrated aluminum frame. Scientific basis is provided for research and development of accuracy and high efficiency of the seed-metering device which is a core component of agricultural machinery of a seeding machine.

Description

Technical Field

[0001] The invention relates to the field of seeding performance detection of a seeding device, and in particular to a method and a device for detecting the seeding performance of a seeding device based on high-frame line scanning. Background Art

[0002] The present invention discloses a method and device for detecting the performance of a precision seeding device based on high-frame line scanning, which is suitable for the performance detection of the seeding device. The detection method of the present invention is to use a linear array camera to perform real-time image acquisition of the falling state of the seed flow during the seeding process, and transmit the image to an image processing system for information processing and analysis. The information processing and analysis includes accurately obtaining the temporal spatial distance of the seeding seed flow through image preprocessing and training based on an improved deep learning network model, analyzing and calculating to obtain the qualified index, missed seeding index, and reseeding index of the seeding device, and other key performance indicators of the seeding device, so as to realize the seeding performance detection of the seeding device; the device used in the present invention includes a stand, a seeding device, an image acquisition device, an image processing device, an inoculation tray, and an integrated aluminum frame. The present invention provides a scientific basis for the research and development of the accuracy and efficiency of the seeding device, which is the core component of the seeding machine agricultural machinery. Summary of the invention

[0003] In view of the above problems, the present invention provides a method and device for detecting the seeding performance of a seed meter based on high-frame line scanning.

[0004] The present invention adopts the following technical solutions:

[0005] A method for online detection of seeding performance of a precision seeding machine based on high frame line scanning comprises the following steps:

[0006] Step 1, install the seeding device to be tested on the stand, set the required working parameters, start the seeding device, and the linear array camera continuously captures images of the horizontal linear field of view directly in front at a preset frame rate, and continuously collects time-series image information of seeds in the seed flow discharged by the seeding device when passing through the linear field of view;

[0007] Step 2: pre-process the valid frames of the collected time-series images; input the images into the trained detection model, and the detection model predicts the pre-processed images and then splices them to obtain the spatial distribution state image of the seed flow. According to the frame rate of the linear array camera and the shooting time of each image, the relative position information and the total number of seeds in the seed flow in the process of falling into the linear field of view are output;

[0008] Step 3: According to the relative position information of various seeds obtained in step 2, the vertical distance between adjacent seeds in the seed stream when they fall into the linear field of view is obtained. Let the vertical distance between the adjacent i-th seed and the i+1-th seed be ΔL i,i+1 , then the actual seeding time interval between the i-th seed and the i+1-th seed falling is Where v2 is the velocity of the seed when it falls into the horizontal linear field of view of the linear array camera, which is a fixed value;

[0009] Step 4: Calculate the theoretical seeding time interval Δt, and compare the theoretical seeding time interval Δt with the actual seeding time interval Δt i,i+1 , judge whether the seeding of the i+1th seed is qualified, missed or re-seeded. After judging all seeds in a batch of seed streams, divide the qualified number, missed number and re-seeded number by the total number of seeds respectively, and calculate the qualified index, missed index and re-seeding index of the seed meter.

[0010] Furthermore, the step 4 specifically includes the following steps:

[0011] Step 4.1: Calculate the theoretical seeding time interval Where n is the rotation speed of the seeding disc, r / min; z is the number of holes;

[0012] Step 4.2: Calculate the actual seeding time interval Δt of the seeding device 实 , compare the actual seeding time interval with the theoretical seeding time interval. If 0.5Δt 理 ≤Δt 实 ≤1.5Δt 理 It is judged as normal sowing, Δt 实 >1.5Δt 理 It is considered as missed broadcast, Δt 实 <0.5Δt 理 It is judged as reseeding, and the number of reseeding, number of missed seeding, and actual seeding number are obtained;

[0013] Step 4.3, qualified index = number of seeds sown normally / total number of seeds, missed sowing index = missed sowing number / total number of seeds, reseeding number = reseeding number / total number of seeds, the qualified index, missed sowing index and reseeding index are calculated based on the seeding information of a batch of seed flows.

[0014] Furthermore, in step 3 Where R is the radius of the seeding disc, n is the rotation speed of the seeding disc, y0 is the distance from the hole just entering the positive pressure zone to the seeding device outlet, and y1 is the vertical distance from the seeding device outlet to the linear array camera lens.

[0015] Furthermore, the method for preprocessing the image in step 1 includes the following steps:

[0016] Step 1.1, converting the obtained grain image into a grayscale image;

[0017] Step 1.2, crop the left and right sides of the grayscale image to reduce the image size;

[0018] Step 1.3, analyze the histogram features of the cropped grayscale image, calculate the first-order forward difference of the number of pixels in the grayscale range of 0 to 10 on the histogram of the original grayscale image, use the grayscale value of the first-order difference maximum point as the grayscale segmentation threshold, and perform binary segmentation on the cropped original grayscale image;

[0019] Step 1.4: Use a circular structure element to perform morphological closing-then-opening smoothing on the binary image to filter edge noise and finally obtain the preprocessed image.

[0020] Furthermore, the method for obtaining the trained detection model includes the following steps:

[0021] Step 2.1, start the seeding device, and the linear array camera continuously captures images of the horizontal linear field of view in front at a preset frame rate, and continuously collects time-series image information of seeds in the seed flow discharged by the seeding device when passing through the linear field of view; filter the collected image valid frames, preprocess the grain image to obtain a preprocessed image, mark and classify the detection targets in the preprocessed image to obtain a data set, the detection targets are single grains and adhered grains, and the data set is classified into one, two, and three, indicating that a single grain, two adhered grains, and three adhered grains can be identified, and the data set is divided into a training set and a validation set in proportion;

[0022] Step 2.2: Use the training set to train the deep learning network model. The model is used to detect targets and output the detection results, which include the anchor frame of the grain, the number of grains and their location information. The optimal model is obtained by minimizing the loss function.

[0023] Step 2.3: Use the optimal model to perform target detection on the validation set to obtain the detection results. Determine whether the error between the detection results and the actual results meets the preset requirements. If yes, use the optimal model as the final detection model. Otherwise, go to step 2.2 to retrain the model.

[0024] Furthermore, the content annotated in step 1 includes the coordinates of the seed and the seed boundary box and the seed category label information, and a corresponding txt file is generated for the annotated content.

[0025] Furthermore, the preprocessed image is an image that meets the input requirements of the deep learning network basic model.

[0026] A device for detecting the seeding performance of a seeding device based on high-frame line scanning, comprising a stand, an image acquisition device, an image processing device, an inoculation tray, and an integrated aluminum frame. The seeding device to be detected can be detachably arranged on the stand, and the image acquisition device and the image processing device are arranged on the integrated aluminum frame.

[0027] The seed meter to be detected is used to discharge seeds in a vertical direction toward the linear array camera below;

[0028] The image acquisition device is used to collect time-series image information of seeds in the seed flow discharged by the seeding device when they pass through the linear field of view, and send the image information to the image processing system;

[0029] The image processing system is used to receive and process and analyze the grain image information collected by the linear array camera, and display the processing results.

[0030] Further, the image acquisition device includes a camera module, a light source module and an integrated aluminum frame;

[0031] The camera module includes an image acquisition card, a camera support frame, a linear array camera, an adapter ring and a camera lens. The image acquisition card is connected to the linear array camera via a CameraLink interface. The camera lens and the linear array camera are connected via threads of the adapter ring. The linear array camera is placed on two beams of the integrated aluminum frame and is fixed in position by the camera support frames on the left and right sides.

[0032] The light source module comprises a strip light source, a light source bracket, an L-shaped adjustable handle and a light source controller; the strip light source is fastened to the light source bracket, and the light source bracket is placed on the beam of the frame and can rotate along the beam at a preset angle.

[0033] The integrated aluminum frame includes an aluminum profile and a connecting plate; the upper part of the integrated aluminum frame is composed of a left and a right top plate, the right top plate is fastened to the aluminum profile, the left top plate is connected to the right top plate by a hinge, one side is fixed and the other side is movable, which is used to adjust the illumination angle of the light source module before image acquisition; a partition is fixed in the middle of the integrated aluminum frame; the light source controller is placed at the bottom of the frame, and a vent is drilled at the bottom of the integrated aluminum frame.

[0034] After adopting the above technical solution, the present invention has the following advantages compared with the prior art:

[0035] 1. Since the theoretical height of the image collected by the linear array camera along the direction of movement is not limited, it can realize the monitoring of grain targets under high-speed dynamic conditions. Conventional image capture can only show the positional relationship between seeds, and it is impossible to determine the order of seeding of different seeds in the same image. The method of the present invention can accurately determine the order and relative position of seed planting, and then determine the seeding performance, thereby providing a new technical reference means and preliminary scientific theoretical basis for seeding performance detection.

[0036] 2. Based on the effective monitoring of the seed flow of small-size seeds such as rapeseed, the system can not only conduct real-time and online seeding uniformity detection of single-seed precision seeders, but also has a certain degree of versatility in monitoring the seeding rate of row seeding seeders.

[0037] 3. A method using a linear array camera to measure the seeding amount of the row seeding device in real time to achieve seeding uniformity detection can more clearly and accurately grasp the impact of the seeding device structure on seeding uniformity, and provide key performance detection equipment for achieving precision seeding.

[0038] 4. A seed counting method based on the YOLO series algorithm of deep learning is proposed to achieve low-cost and high-accuracy automatic grain counting.

[0039] 5. Simple structure, easy to operate, convenient for performance testing of the seed meter.

[0040] The present invention is described in detail below with reference to the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 It is a schematic diagram of a seeding performance detection device of a seeding device based on high frame line scanning of the present invention;

[0042] Figure 2 for Figure 1 Schematic diagram of the structure of the middle seed discharge device;

[0043] Figure 3 for Figure 1 Schematic diagram of the structure of the detection device;

[0044] Figure 4 for Figure 3 Schematic diagram of the camera module;

[0045] Figure 5 for Figure 3 Schematic diagram of the central light source module;

[0046] Figure 6 It is a system workflow diagram of the present invention;

[0047] Figure 7 It is the principle diagram of the detection method of the present invention;

[0048] Figure 8 This is a visualization of the training results.

[0049] The parts represented by each number are as follows:

[0050] 1. Stand; 2. Seeding device; 3. Detection device; 4. Human-computer interaction interface; 201. Seeding device positive pressure pipeline; 202. Motor frame; 203. Stepper motor; 204. Elastic coupling; 205. Seeding device; 206. Seeding device negative pressure pipeline; 207. AC speed regulating fan; 301. Integrated aluminum frame; 302. Light source controller; 303. Chassis; 304. Partition; 305. Camera module; 306. Light source module; 307A. Left top plate; 307B. Right top plate; 308. Hinge; 309. Inoculation tray; 310. Connecting plate; 301A. Camera support frame; 301B. Linear array camera; 301C. Connecting ring; 301D. Camera lens; 306A. L-shaped adjustable handle; 306B. Light source bracket; 306C. Strip light source. DETAILED DESCRIPTION

[0051] The principles and features of the present invention are described below in conjunction with the accompanying drawings. The examples given are only used to explain the present invention and are not used to limit the scope of the present invention.

[0052] like Figure 1-5 As shown, a method and device for detecting the seeding performance of a precision seed meter based on high-frame line scanning include a stand 1, a seed metering device 2, an image acquisition device, an image processing device, an inoculation tray 309, and an integrated aluminum frame 301; the image acquisition device, the image processing device, the integrated aluminum frame, and the connecting plate together constitute a detection device 3.

[0053] Combination Figure 1 The platform is assembled with 4040 aluminum profiles and fixed in the middle by hinges.

[0054] Combination Figure 2 The seeding device 2 includes a seeding device 205, a fan 207, an elastic coupling 204, a stepping motor 203, a positive pressure pipeline 201 and a negative pressure pipeline 206. The seeding device 205 is a pneumatic seeding device. The fan 207 is used to provide positive pressure and negative pressure to the seeding device 205, and cooperate with the seeding device 205 to perform seeding. The elastic coupling 204 is assembled on the stepping motor 203 and connected to the rotating shaft of the seeding device 205. The seeding device 2 is detachably arranged on the stand 1; the seeding device 2 is used to supply seeds to the image acquisition device.

[0055] Combination Figure 3The detection device 3 includes an image acquisition device, an image processing device, an integrated aluminum frame 301, a connecting plate 310, an upper top plate 307, and a partition 304. The image acquisition device includes a camera module 305, a light source module 306, and a light source controller 302. The image processing device includes a chassis 303 and a human-computer interaction interface 4. The detection device 3 is mounted on the stand 1 through two aluminum profiles below, and the seed drop height can be changed by changing the distance of the aluminum profile from the ground. The human-computer interaction interface 4 is embedded in the rectangular cut of the left connecting plate of the detection device 3 and is connected to the chassis with a connecting line.

[0056] As an implementation mode, the rectangular hole of the right top plate 307B above the detection device 3 is directly opposite to the seed drop port of the seed metering device 205. The rectangular window with a cutting size of 190×100mm is convenient for seed drop and grain image shooting in conjunction with the shooting field of view of the industrial camera 301B and the working distance of the lens 301D. An inoculation tray 309 is provided at the bottom of the device to facilitate the recovery and utilization of grains. The device for detecting the performance of the seed metering device based on high-frame line scanning is characterized in that the industrial camera 301B is a linear array scanning element. The linear array scanning element performs line-by-line scanning due to the relative motion between the camera and the object to be detected. Through high-frequency scanning, each line image of the object to be detected is collected, and finally these linear images are output, and a planar array image is re-stacked through the order of collection. By setting the scanning frequency of the industrial camera 301B, the grain target detection under high-speed state can be successfully achieved.

[0057] As an implementation mode, four through holes with a diameter of 6 mm are drilled on the right top plate 307B of the detection device 3, which can be fastened to the rectangular material by bolts and nuts, and the left top plate 307A and the right top plate 307B are connected by hinges 308 to realize the opening and closing of one side of the left top plate 307A. Before image acquisition, the position and angle of the light source 306C can be adjusted by opening the left top plate 307A, and the top plate can be closed to isolate the external environment interference when image acquisition is performed.

[0058] As an implementation mode, an inoculation tray 309 is placed at the bottom of the detection device 3, and the inoculation tray 309 is located directly below the rectangular window of the right top plate 307B.

[0059] Combination Figure 4, the camera module 305 includes a camera support frame 301A, an industrial camera 301B, an adapter ring 301C, and a camera lens 301D. The industrial camera 301B uses the Hikvision MV-CL082-92CM model, with a resolution of 8192×2 and a maximum line frequency of 100KHz. The camera lens 301D uses the LS8003A model. The industrial camera 301B and the camera lens 301D are fixed by a threaded connection of the adapter ring 301C. The camera support frame 301A is arranged on both sides of the industrial camera 301B and is fixed with screws. The camera support frame has the same height as the linear array camera. Two through holes with a diameter of 5mm are drilled on both the upper and lower sides. It is fastened in the middle of the two beams of the detection device with bolts and nuts to limit the movement of the camera; the partition 304 is placed in the middle of the detection device and fixed to the middle aluminum profile with a T-bolt pair.

[0060] Combination Figure 5 The light source module 306 includes two symmetrical L-shaped adjustable handles 306A, a light source bracket 306B, a strip light source 306C and a light source controller 302; the L-shaped adjustable handles 306A and bolts and nuts can fix the light source bracket 306B on the aluminum profiles on the left and right sides. Both aluminum profiles are drilled with 6 threaded holes with a diameter of 6mm and a hole spacing of 15mm, which are used to adjust the horizontal position and angle of the strip light source 306C irradiating the grains. A 100mm long through groove is drilled on the outer connecting plate of the aluminum profile, which is used to fix the light source bracket in conjunction with the aluminum profile.

[0061] The strip light source 306C uses the ultra-bright strip light source QLH-22516, and is fixed to the light source bracket 306B by bolts and nuts. The adjustable angle range of the strip light source 306C is 0-90°.

[0062] Specific implementation process:

[0063] 1) Set the required working parameters, start the seeding device, the seeding device stepper motor and AC speed-adjustable fan start working, collect the image information of each seed when it passes through the detection device, filter the valid frames of the collected images, pre-process the seed images to obtain the pre-processed images, use the Labelimg tool to mark and classify the detection targets in the pre-processed images, and obtain the data set;

[0064] The detection targets are single grains and adhered grains;

[0065] The data set is divided into a training set and a validation set in a ratio of 8:2;

[0066] 2) Based on the YOLOV7 network basic model, construct and add the BiFormer module, introduce the Focal Loss loss function, use Cutout on Mosaic, and use the EfficientNet lightweight feature extraction network to compress the network configuration to obtain the YOLOV7 network improved model;

[0067] 3) Use the training set to iterate the training of the YOLOV7 network improved model to obtain the optimal model; use the optimized model to perform target detection on the validation set to obtain the detection results, which include the seed anchor frame and the number of seeds and their location information;

[0068] 4) Combination Figure 7 , according to the number of grains and the parameter information between grains contained in each frame of grayscale image in the seed flow obtained in step 3, the performance parameters of the seed meter are calculated: by calculating the vertical component velocity v1 of the seed leaving the seed drop port, as well as the seed disk speed n and the seed disk radius R and other parameters, the instantaneous velocity v2 of the seed when it leaves the straight line segment y1 and is scanned by the linear array camera can be obtained. The instantaneous velocity v2 of the grain when it enters the field of view of the industrial camera and the vertical physical spacing L between adjacent seeds processed by the grain position information in the output image of the linear array camera can be calculated to obtain the actual seeding time interval The specific calculation formula is as follows:

[0069]

[0070] Where n is the rotation speed of the seeding disc, r / min; z is the number of holes;

[0071] Calculate the actual seeding time interval △t when adjacent seeds in the seed flow leave the seed guide port and enter the detection device 实 , where L is the actual physical distance between the centroids of adjacent seeds when they pass through the field of view of the linear array camera and are captured in the output image one after another, v1 is the vertical component velocity of the seed leaving the seed guide port, v2 is the seed velocity captured by the linear array camera when the seed drops a length of y1 from the seed guide port, y0 is the distance between the shaped hole and the seed meter outlet, and y1 is the vertical distance between the seed drop port and the camera lens;

[0072] By comparing with the theoretical seeding time interval, the seeding performance index after the seed flow leaves the seeding device is obtained, and the seeding performance index includes the qualified index, the missed seeding index and the reseeding index;

[0073] Combination Figure 8 , a is the first seed that passes through the linear field of view of the linear array camera, i.e., the first seed captured by the camera, and the next seed b passes through Δt 实 The time is captured by the camera. The vertical distance L1 between the two grains in the output image is the number of pixels of the adjacent seed centers in the vertical direction multiplied by the physical size of each pixel, which is also the time it takes for the two grains to pass Δt at a speed of v2.实 The distance scanned in sequence. Within a certain period of time, seeds c, d, and e passed through the linear field of view of the linear array camera in sequence and were successfully captured. In the figure, the colored shading next to the grains represents the classification and confidence of the prediction results by the model. The confidence is usually represented by a value between 0 and 1, that is, the probability that grain a is One is 0.91, the probability that grain b is Two is 0.82, the probability that grain c is One is 0.90, the probability that grain d is One is 0.94, and the probability that grain e is One is 0.86. The actual seeding time interval Δt between the two adjacent grains in the image in the vertical direction is calculated by the number of pixels in the image and the instantaneous speed of the grains when they are scanned by the linear array camera. 实 , compare the actual physical distances between adjacent grains in the vertical direction to determine whether there is any reseeding or missing between adjacent grains.

[0074] As an implementation manner, in step 4, the method for calculating the qualified index, the missed broadcast index and the replay index comprises the following steps:

[0075] 2.1) Calculate the theoretical seeding time interval Where n is the rotation speed of the seeding disc, r / min; z is the number of holes;

[0076] 2.2) Calculate the actual seeding time interval Δt between adjacent seeds in the seed flow leaving the seed guide port and entering the field of view of the linear array camera 实 , compare the actual seeding time interval with the theoretical seeding time interval. If 0.5Δt 理 ≤Δt 实 ≤1.5Δt 理 It is judged as normal sowing, Δt 实 >1.5Δt 理 It is considered as missed broadcast, Δt 实 <0.5Δt 理 It is judged as a replay;

[0077] 2.3) The test uses the line scanning camera at the seed drop port and below the seed drop port to detect the time data of the seeding process, collect the spatial distribution state image of the seed flow, continuously monitor the seeding time interval of 251 seeds, and output the data of this batch of seed flows to calculate the key performance indicators of the seeding device such as the qualified index, missed seeding index and reseeding index.

[0078] The above is an example of the best implementation of the present invention, and the parts not described in detail are common knowledge of ordinary technicians in the field. The protection scope of the present invention shall be based on the content of the claims, and any equivalent transformation based on the technical enlightenment of the present invention is also within the protection scope of the present invention.

Claims

1. A method for online detection of seeding performance of a seeding device based on high frame line scanning, characterized in that: The following steps are involved: Step 1, install the seeding device to be tested on the stand, set the required working parameters, start the seeding device, and the linear array camera continuously captures images of the horizontal linear field of view directly in front at a preset frame rate, and continuously collects time-series image information of seeds in the seed flow discharged by the seeding device when passing through the linear field of view; Step 2, preprocess the collected time-series image frames; input the images into the trained detection model, the detection model predicts the preprocessed images and then splices them to obtain the spatial distribution state image of the seed flow, and outputs the relative position information and the total number of seeds in the seed flow in the process of falling into the linear field of view according to the frame rate of the linear array camera and the shooting time of each image; Step 3: According to the relative position information of various seeds obtained in step 2, the vertical distance between adjacent seeds in the seed stream when they fall into the linear field of view is obtained. Let the vertical distance between the adjacent i-th seed and i+1-th seed be ΔL i,i+1 , then the actual seeding time interval between the i-th seed and the i+1-th seed falling is Where v2 is the velocity of the seed when it falls into the horizontal linear field of view of the linear array camera, which is a fixed value; Step 4: Calculate the theoretical seeding time interval Δt, and compare the theoretical seeding time interval Δt with the actual seeding time interval Δt i,i+1 , judge whether the seeding of the i+1th seed is qualified, missed or re-seeded. After judging all seeds in a batch of seed streams, divide the qualified number, missed number and re-seeded number by the total number of seeds respectively, and calculate the qualified index, missed index and re-seeding index of the seed meter.

2. The method for detecting performance of a seed meter based on high frame line scanning according to claim 1, characterized in that: The step 4 specifically comprises the following steps: Step 4.1: Calculate the theoretical seeding time interval Where n is the rotation speed of the seeding disc, r / min; z is the number of holes; Step 4.2: Calculate the actual seeding time interval Δt of the seeding device 实 , compare the actual seeding time interval with the theoretical seeding time interval. If 0.5Δt 理 ≤Δt 实 ≤1.5Δt 理 It is judged as normal sowing, Δt 实 >1.5Δt 理 It is considered as missed broadcast, Δt 实 <0.5Δt 理 It is judged as reseeding, and the number of reseeding, number of missed seeding, and actual seeding number are obtained; Step 4.3, qualified index = number of seeds sown normally / total number of seeds, missed sowing index = missed sowing number / total number of seeds, reseeding number = reseeding number / total number of seeds, the qualified index, missed sowing index and reseeding index are calculated based on the seeding information of a batch of seed flows.

3. The method for detecting performance of a seed meter based on high frame line scanning according to claim 1, characterized in that: In step 3 Where R is the radius of the seeding disc, n is the rotation speed of the seeding disc, y0 is the distance from the hole just entering the positive pressure zone to the seeding device outlet, and y1 is the vertical distance from the seeding device outlet to the linear array camera lens.

4. The method for detecting performance of a seed meter based on high frame line scanning according to claim 1, characterized in that: The method for preprocessing the image in step 1 includes the following steps: Step 1.1, converting the obtained grain image into a grayscale image; Step 1.2, crop the left and right sides of the grayscale image to reduce the image size; Step 1.3, analyze the histogram features of the cropped grayscale image, calculate the first-order forward difference of the number of pixels in the grayscale range of 0 to 10 on the histogram of the original grayscale image, use the grayscale value of the first-order difference maximum point as the grayscale segmentation threshold, and perform binary segmentation on the cropped original grayscale image; Step 1.4: Use a circular structure element to perform morphological closing-then-opening smoothing on the binary image to filter edge noise and finally obtain the preprocessed image.

5. The method for detecting performance of a seed meter based on high frame line scanning according to claim 1, characterized in that: The method for obtaining a trained detection model includes the following steps: Step 2.1, start the seeding device, and the linear array camera continuously captures images of the horizontal linear field of view in front at a preset frame rate, and continuously collects time-series image information of seeds in the seed flow discharged by the seeding device when passing through the linear field of view; filter the collected image valid frames, preprocess the grain image to obtain a preprocessed image, mark and classify the detection targets in the preprocessed image to obtain a data set, the detection targets are single grains and adhered grains, and the data set is classified into one, two, and three, indicating that a single grain, two adhered grains, and three adhered grains can be identified, and the data set is divided into a training set and a validation set in proportion; Step 2.2: Use the training set to train the deep learning network model. The model is used to detect targets and output the detection results, which include the anchor frame of the grain, the number of grains and their location information. The optimal model is obtained by minimizing the loss function. Step 2.3: Use the optimal model to perform target detection on the validation set to obtain the detection results. Determine whether the error between the detection results and the actual results meets the preset requirements. If yes, use the optimal model as the final detection model. Otherwise, go to step 2.2 to retrain the model.

6. The method for detecting seeding performance of a seed meter based on high frame line scanning according to claim 1, characterized in that: The content marked in step 1 includes the coordinates of the seed and the seed boundary box and the seed category label information, and a corresponding txt file is generated for the marked content.

7. The method for detecting performance of a seed meter based on high frame line scanning according to claim 1, characterized in that: The preprocessed images are images that meet the input requirements of the deep learning network basic model.

8. A device for detecting the seeding performance of a seed meter based on high frame line scanning, characterized in that: It includes a stand, a seeding device, an image acquisition device and an image processing device. The seeding device to be detected can be detachably arranged on the stand, and the image acquisition device and the image processing device are arranged on an integrated aluminum frame. The seed meter to be detected is used to discharge seeds in a vertical direction toward the linear array camera below; The image acquisition device is used to collect time-series image information of seeds in the seed flow discharged by the seeding device when they pass through the linear field of view, and send the image information to the image processing device; The image processing device is used to receive and process and analyze the grain image collected by the linear array camera, and display the processing result.

9. The device for detecting the seeding performance of a seed meter based on high frame line scanning according to claim, characterized in that: The image acquisition device includes a camera module, a light source module and an integrated aluminum frame; The camera module includes an image acquisition card, a camera support frame, a linear array camera, an adapter ring and a camera lens. The image acquisition card is connected to the linear array camera via a CameraLink interface. The camera lens and the linear array camera are connected via threads of the adapter ring. The linear array camera is placed on two beams of the integrated aluminum frame and is fixed in position by the camera support frames on the left and right sides. The light source module comprises a strip light source, a light source bracket, an L-shaped adjustable handle and a light source controller; the strip light source is fastened to the light source bracket, and the light source bracket is placed on the beam of the integrated aluminum frame and can rotate along the beam at a preset angle. The integrated aluminum frame includes an aluminum profile and a connecting plate; the upper part of the integrated aluminum frame is composed of a left and a right top plate, the right top plate is fastened to the aluminum profile, the left top plate is connected to the right top plate by a hinge, one side is fixed and the other side is movable, which is used to adjust the illumination angle of the light source module before image acquisition; a partition is fixed in the middle of the integrated aluminum frame; the light source controller is placed at the bottom of the frame, and a vent is drilled at the bottom of the integrated aluminum frame.

Images