High-throughput seed sorting system based on deep learning

Abstract

The invention provides a high-throughput seed sorting system based on deep learning, and belongs to the field of agricultural electrical automation equipment.The high-throughput seed sorting system comprises a feeding module, an identification module and a sorting module, the identification module is arranged on one side of the feeding module, and the sorting module is arranged on one side of the identification module; the recognition module is in communication connection with the sorting module through an upper computer, the feeding module is provided with a feeding hopper and a two-stage electromagnetic vibration feeder, the feeding hopper is arranged above the two-stage electromagnetic vibration feeder, a conveying belt is arranged at the tail end of the two-stage electromagnetic vibration feeder, the recognition module is arranged above the conveying belt, and the sorting module is arranged above the conveying belt. A sorting module is arranged at the tail end of the conveying belt. By adopting the high-throughput seed sorting system based on deep learning, high-throughput and high-efficiency sorting of seeds is realized.

Description

Technical Field

[0001] The present invention relates to the field of agricultural electrical automation equipment, and in particular to a high-throughput seed sorting system based on deep learning. Background Art

[0002] Seeds, as the cornerstone of agricultural production, their quality is directly related to food security and the vital interests of farmers. High-quality seeds are the core "chips" that drive agricultural modernization and enhance the competitiveness of the agricultural industry. With the increasing popularity of machine vision technology, compared with traditional manual and mechanical sorting methods, this technology has shown significant advantages. It can not only effectively avoid visual fatigue and subjective judgment deviation in manual operations, but also greatly make up for the deficiencies of pure mechanical sorting in identifying damaged seeds such as insect-eaten, diseased spots, and breakages.

[0003] Therefore, the research and development of an automatic sorting device that integrates computer vision technology, is specifically targeted at small target seeds, and has the characteristics of high throughput and high efficiency has become an urgent task for current agricultural technological innovation. The high-throughput seed automatic sorting system, with its dual advantages of high precision and high speed, is particularly suitable for accurately identifying and efficiently separating multiple varieties of seeds in a complex background. This design and technological innovation provide strong technical support for improving agricultural production efficiency and ensuring seed quality, indicating that it will have broad application prospects in the agricultural field. Summary of the Invention

[0004] The purpose of the present invention is to provide a high-throughput seed sorting system based on deep learning, which realizes high-throughput and high-efficiency sorting of seeds.

[0005] To achieve the above object, the present invention provides a high-throughput seed sorting system based on deep learning, including a feeding module, an identification module, and a sorting module. The identification module is arranged on one side of the feeding module, and the sorting module is arranged on one side of the identification module. The identification module is communicatively connected to the sorting module through a host computer. The feeding module is provided with a feeding hopper and a two-stage electromagnetic vibration feeder. The feeding hopper is arranged above the two-stage electromagnetic vibration feeder. The end of the two-stage electromagnetic vibration feeder is provided with a conveyor belt. The identification module is arranged above the conveyor belt, and the end of the conveyor belt is provided with a sorting module.

[0006] Preferably, a seed type carrier is arranged above the two-stage electromagnetic vibration feeder. The seed type carrier is provided with folding grooves, and the center point spacing between adjacent grooves of the folding grooves is the same.

[0007] Preferably, the recognition module includes a seed recognition unit and a bad seed information sending unit. An industrial line scan camera is provided on the seed recognition unit, and a target detection model is carried on the industrial line scan camera. The industrial line scan camera is communicatively connected to a host computer, and the bad seed information sending unit is communicatively connected to the host computer.

[0008] Preferably, the information sent by the bad seed information sending unit includes the coordinates of the bad seeds. The bad seed coordinates are (X, C), where X is the distance of the bad seed from the end of the conveyor belt, and C is the channel information of the bad seed.

[0009] Preferably, the sorting module includes a control unit and a blowing and sorting unit. The control unit is communicatively connected to the host computer. The control unit receives the bad seed coordinates through the host computer, and the control unit is electrically connected to the blowing and sorting unit.

[0010] Preferably, a channel air blowing port corresponding to the folding groove is provided on the blowing and sorting unit, and the number of the channel air blowing ports is the same as the number of the folding grooves.

[0011] Preferably, a servo motor and a synchronous pulley are provided on one side of the conveyor belt. The output shaft of the servo motor is connected to the synchronous pulley, and the synchronous pulley is engaged with the conveyor belt in a toothed manner.

[0012] Preferably, the blowing and sorting unit is provided with a blowing valve plate, and a sorting funnel is provided below the blowing valve plate. The sorting funnel corresponds to the channel air blowing port.

[0013] Therefore, the present invention adopts the above-mentioned high-throughput seed sorting system based on deep learning, and the technical effects are as follows:

[0014] 1. High-throughput processing: Two-stage electromagnetic vibration feeders are adopted, which can stably and continuously supply seeds, ensuring the uninterrupted progress of the sorting process; the design of the folding grooves on the seed carrier enables the seeds to be arranged in an orderly manner and pass through the recognition module, avoiding the accumulation and chaos of seeds and improving the processing efficiency.

[0015] 2. High-efficiency recognition and sorting: The industrial line scan camera is equipped with a target detection model, which can identify bad seeds in real time and accurately, and quickly send the bad seed information (including coordinates) to the host computer; the host computer timely transmits the bad seed coordinates to the control unit of the sorting module through the communication connection, realizing the close connection between recognition and sorting; the blowing and sorting unit blows air precisely through the corresponding channel air blowing ports according to the received bad seed coordinates, quickly and accurately separating the bad seeds, avoiding manual intervention and further improving the sorting efficiency. Description of the Drawings

[0016] Figure 1Schematic diagram of the overall seed sorting device of the present invention;

[0017] Figure 2 Top view of the seed sorting device of the present invention;

[0018] Figure 3 Schematic diagram of the control processing double queue;

[0019] Figure 4 Schematic diagram of the indicator light of the present invention;

[0020] Figure 5 Experimental effect diagram of the embodiment of the present invention; Figure 5 (a) Abnormal seed image screened by channel 1; Figure 5 (b) Abnormal seed image screened by channel 2; Figure 5 (c) Abnormal seed image screened by channel 3; Figure 5 (d) Abnormal seed image screened by channel 4; Figure 5 (e) Abnormal seed image screened by channel 5; Figure 5 (f) Abnormal seed image screened by channel 6; Figure 5 (g) Abnormal seed image screened by channel 7; Figure 5 (h) Abnormal seed image screened by channel 8.

[0021] Reference numerals

[0022] 1. Feeding hopper; 2. Two-stage electromagnetic vibration feeder; 3. Conveyor belt; 4. Folding groove; 5. Industrial line scan camera; 6. Control panel; 7. Channel air blowing port; 8. Synchronous pulley. Detailed implementation manners

[0023] The technical solutions of the present invention will be further described below with reference to the drawings and embodiments.

[0024] Unless otherwise defined, the technical terms or scientific terms used in the present invention shall have the ordinary meanings understood by those of ordinary skill in the field to which the present invention belongs.

[0025] Embodiment 1

[0026] The present invention provides a high-throughput seed sorting system based on deep learning, which mainly consists of three core parts: a feeding module, an identification module, and a sorting module. The entire system is based on a scientific mechanical structure design, integrating advanced visual recognition technology and automation control technology, aiming to achieve efficient and non-destructive sorting of seeds.

[0027] As Figures 1-2As shown in the figure, the feeding module is located at the starting end of the system, and its main function is to realize the automatic feeding of the seeds to be sorted. This module includes a feeding hopper 1 and a two-stage electromagnetic vibrating feeder 2. The feeding hopper 1 is arranged above the two-stage electromagnetic vibrating feeder 2 and is used to accommodate the seeds to be sorted. Through the vibrating action, the two-stage electromagnetic vibrating feeder 2 disperses and arranges the piled-up seeds into a linear state, thus realizing orderly feeding. In addition, a seed carrier is arranged above the two-stage electromagnetic vibrating feeder 2, and the seed carrier is provided with folding grooves 4. The adjacent center point distances of these grooves are the same, ensuring that the seeds are evenly distributed on the conveyor belt 3 and there is an appropriate gap between the seeds in a single channel, avoiding the mis-blowing phenomenon caused by too small a seed spacing.

[0028] The feeding module part adopts a cabinet structure, which is welded by a base, a square feeding hopper 1, a two-stage electromagnetic vibrating feeder 2, and corresponding aluminum profile brackets. The seed carrier is made of an aluminum plate through folding and welding processes. The adjacent groove center point distance is 18.8 mm. By adjusting the vibration intensity knob on the control panel 6, the vibration intensity can be adaptively adjusted according to different conveyor belt 3 speeds, ensuring that the seeds are evenly distributed on the conveyor belt 3 and there is an appropriate gap between the seeds in a single channel.

[0029] Immediately following the feeding module is the identification module, which is responsible for the image acquisition of the seeds and the identification of bad seeds. The identification module includes a seed identification unit and a bad seed information sending unit. An industrial line scan camera 5 is installed on the seed identification unit. This camera is equipped with an improved version of the YOLOv11 target detection model, which can detect and accurately identify bad seeds on the conveyor belt 3 in real time. The industrial line scan camera 5 is communicatively connected to the host computer, and uploads the collected image data to the host computer for processing. The host computer sorts the identification results into the form of a data string. As Figure 3 shown in the figure, its content includes the distance of the bad seed from the end of the conveyor belt 3 and the corresponding channel information (X, C). The sorted data string is sent to the industrial control board with ARM Cortex-M4 as the core through the communication interface for subsequent sorting control. The bad seed information sending unit is also communicatively connected to the host computer and is responsible for sending the bad seed coordinate information processed by the host computer to the sorting module to achieve precise sorting. The bad seed coordinate information includes the distance (X) of the bad seed from the end of the conveyor belt 3 and the corresponding channel information (C).

[0030] The sorting module is the execution part of the system. According to the bad seed coordinate information sent by the recognition module, it precisely removes the bad seeds. The sorting module includes a control unit and a blowing sorting unit. The control unit is an industrial control board with an ARM Cortex-M4 as the core, communicating with the host computer. It receives and processes the bad seed coordinate information sent by the host computer. The control unit also connects to an incremental rotary encoder that rotates synchronously with the conveyor belt 3 through an external interrupt, and processes the coordinate information based on the positioning pulses of the encoder. The blowing sorting unit then blows and removes the bad seeds according to the instructions of the control unit. There are channel blowing ports 7 on the blowing sorting unit corresponding to the folding grooves 4. The number of these blowing ports is the same as the number of folding grooves 4 to ensure that the bad seeds in each channel can be precisely removed. A sorting funnel is also provided below the blowing sorting unit to collect the removed bad seeds.

[0031] The sorting control module is located above the channel blowing ports 7, and an LED light indication unit is designed corresponding to the channels. As Figure 4 shown, this unit is welded on a PCB board by an STC15 single-chip microcomputer chip, LED light-emitting diodes, a triode amplification circuit, and two corresponding capacitors and other components. After the whole machine is spray-painted and encapsulated, it can realize the function that when a bad seed is blown and removed from a certain channel, the corresponding indicator light flashes.

[0032] The seed recognition unit collects a data set through an industrial line-scan camera 5. The experimental samples consist of 317 chili seeds with reddish-brown or mildew spots (caused by a humid environment) and 683 healthy light-yellow chili seeds, totaling 1000 seeds. Simulating the actual sorting scenario, the feeding unit is used to make the seeds form an eight-channel seed flow through the folding grooves 4, and images are collected to construct a data set containing 1247 pictures, which are divided into a training set, a test set, and a validation set according to a ratio of 8:1.5:0.5. The bad seeds are calibrated using the labelimg tool, and it is trained for 200 rounds on a computer equipped with an Nvidia GeForce RTX3090 graphics card. The improved YOLOv11 is integrated with the Mamba module to accelerate neural network inference and optimize the computational graph. The bad seed coordinate information sending unit is implemented through the host computer. Using the Python language and based on the pyserial module to configure the physical layer serial communication parameters, including the serial port address, baud rate, data bits, stop bits, and timeout waiting time. Users can adjust the parameters in the code according to the actual hardware connection situation. This part adopts a multi-threaded parallel processing mechanism, respectively creating and running a camera frame capture thread, a signal listening thread, and a frame processing thread to improve the system efficiency; as Figure 5 shown, the screen real-time display interface is written using the pyqt module in the Python language. Each channel is provided with an independent visualization window, and a total of 8 windows correspond to the real-time bad seed images of 8 channels. Figure 5(a) shows the abnormal seed image screened by Channel 1. The coordinates of the abnormal seed are (155, 91), and the area is 2; Figure 5 (b) shows the abnormal seed image screened by Channel 2. The coordinates of the abnormal seed are (511, 103), and the area is 4; Figure 5 (c) shows the abnormal seed image screened by Channel 3. The coordinates of the abnormal seed are (311, 123), and the area is 3; Figure 5 (d) shows the abnormal seed image screened by Channel 4. The coordinates of the abnormal seed are (259, 161), and the area is 4; Figure 5 (e) shows the abnormal seed image screened by Channel 5. The coordinates of the abnormal seed are (445, 236), and the area is 3; Figure 5 (f) shows the abnormal seed image screened by Channel 6. The coordinates of the abnormal seed are (366, 265), and the area is 5; Figure 5 (g) shows the abnormal seed image screened by Channel 7. The coordinates of the abnormal seed are (386, 311), and the area is 5; Figure 5 (h) shows the abnormal seed image screened by Channel 8. The coordinates of the abnormal seed are (257, 332), and the area is 3. The sorting status of the seeds is displayed in real time. When a bad seed to be removed is detected, the seed will be highlighted in the corresponding window, and its coordinate position, area size, and the cumulative number of bad seeds passed through the current channel will be recorded synchronously, providing comprehensive quality monitoring and mass production statistics support for users.

[0033] In addition, to ensure the stable operation of conveyor belt 3, a servo motor and a synchronous pulley 8 are also set on one side of conveyor belt 3. The output shaft of the servo motor is connected to synchronous pulley 8, and synchronous pulley 8 is connected to conveyor belt 3 in a toothed engagement manner, thereby driving conveyor belt 3 to rotate at a constant speed of 30 cm / s.

[0034] During the actual operation process, the sorting function of the upper computer is adjusted and enabled, and the servo motor and the two-stage electromagnetic vibration feeder 2 are started at the control panel 6. Subsequently, the operation of feeding the seeds to be sorted is carried out at the hopper 1. At this time, the piled-up seeds will pass through the folding groove 4 above the electromagnetic vibration feeder, and its unique folding ripple design can make the seeds gradually form a linear material source. Then, the industrial line scan camera 5 will capture and analyze the seed images in real time, and at the same time, the industrial control board with ARM Cortex-M4 as the core will process the analysis results. Finally, the system can accurately identify the bad seeds and remove them at the air-blowing sorting structure.

[0035] Therefore, the present invention adopts the above-mentioned high-throughput seed sorting system based on deep learning, which integrates the latest deep learning visual recognition technology and precise automatic control technology, achieving high-throughput, real-time, non-destructive and accurate seed sorting. Compared with the existing color sorters and seed sorting systems, the present invention has higher recognition accuracy and frame rate, can monitor the high-throughput seed flow on the conveyor belt in real time at a stable high frame rate, and accurately and quickly capture the contour, lesions, color texture and other characteristic information of the seeds to be removed, improving the efficiency and accuracy of seed sorting.

[0036] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that they can still modify or equivalently replace the technical solutions of the present invention, and these modifications or equivalent replacements cannot make the modified technical solutions deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A high-throughput seed sorting system based on deep learning, characterized in that, It includes a feeding module, an identification module and a sorting module. The identification module is arranged on one side of the feeding module, and the sorting module is arranged on one side of the identification module. The identification module is communicatively connected to the sorting module through a host computer. The feeding module is provided with a feeding hopper and a two-stage electromagnetic vibrating feeder. The feeding hopper is arranged above the two-stage electromagnetic vibrating feeder. The end of the two-stage electromagnetic vibrating feeder is provided with a conveyor belt. The identification module is arranged above the conveyor belt, and the sorting module is arranged at the end of the conveyor belt.

2. The high-throughput seed sorting system based on deep learning according to claim 1, wherein A seed type carrier is arranged above the two-stage electromagnetic vibrating feeder. The seed type carrier is provided with folding grooves, and the center point spacing between adjacent grooves of the folding grooves is the same.

3. The high-throughput seed sorting system based on deep learning according to claim 1, characterized in that, The identification module includes a seed identification unit and a bad seed information sending unit. An industrial line scanning camera is arranged on the seed identification unit. A target detection model is carried on the industrial line scanning camera. The industrial line scanning camera is communicatively connected to the host computer, and the bad seed information sending unit is communicatively connected to the host computer.

4. The high-throughput seed sorting system based on deep learning according to claim 3, wherein, The information sent by the bad seed information sending unit includes the coordinates of bad seeds. The coordinates of bad seeds are (X, C), where X is the distance of bad seeds from the end of the conveyor belt, and C is the channel information of bad seeds.

5. The high-throughput seed sorting system based on deep learning according to claim 1, wherein The sorting module includes a control unit and a blowing and sorting unit. The control unit is communicatively connected to the host computer. The control unit receives the coordinates of bad seeds through the host computer, and the control unit is electrically connected to the blowing and sorting unit.

6. The high-throughput seed sorting system based on deep learning according to claim 5, characterized in that, Channel blowing ports corresponding to the folding grooves are formed on the blowing and sorting unit, and the number of the channel blowing ports is the same as that of the folding grooves.

7. A high-throughput seed sorting system based on deep learning according to claim 1, characterized in that, A servo motor and a synchronous pulley are arranged on one side of the conveyor belt. The output shaft of the servo motor is connected to the synchronous pulley, and the synchronous pulley is in tooth-shaped engagement with the conveyor belt.

8. The high-throughput seed sorting system based on deep learning according to claim 6, wherein The blowing and sorting unit is provided with a blowing valve plate. A sorting funnel is arranged below the blowing valve plate, and the sorting funnel corresponds to the channel blowing ports.

Images