Granular crop grain-by-grain ordering device

Abstract

The granular crop grain-by-grain orderly arrangement device comprises sequentially arranged orderly arrangement mechanism, vibration feeding mechanism, material conveying mechanism and image acquisition module, and control system connected with vibration feeding mechanism and image acquisition module respectively, and the granular crop can be arranged grain by grain automatically and orderly in a consistent posture in feeding and conveying, laid flat in the imaging field of view of the image acquisition module, the visual features of the target to be detected in the image are fully exposed, the speed and accuracy of image segmentation, identification and classification are increased, the identification accuracy of imperfect grains, weeds, insects and impurities in the raw grains is improved, and therefore the granular crop is automatically, quickly and accurately detected and classified by machine vision.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of crop sorting, and particularly relates to a granular crop (such as grain crops such as wheat, soybeans, corn, etc., or economic crops such as peanuts, rapeseed, coffee beans, etc.) granule orderly arrangement device. BACKGROUND

[0002] Precise granule detection of granular crops is a necessary measure to ensure grain quality and maintain food safety. Current granule detection methods mainly involve uniformly spreading crop samples on a test table by quality inspection personnel, observing and picking out imperfect grains, weeds, insects and impurities one by one by hand using a magnifying glass, and identifying them according to their own professional knowledge and in combination with relevant national standards. Some samples with difficult-to-distinguish features also need to be moved under a microscope for fine observation or expert consultation. Therefore, the above method has many limitations, such as long time consumption for manually spreading and granule detection of a large number of crops, and subjective errors that will have a predetermined impact on accuracy. It is far from meeting the requirements of automatic, intelligent and high-precision granule detection of granular crops on site at ports and in grain stores. SUMMARY

[0003] The present application proposes a granular crop granule orderly arrangement device to address the problem that the existing technology does not consider the size difference of individual particles or the blockage of a single chute caused by the mixing of different types of particles, and the random movement of grain particles caused by the inertial effect of falling from the inclined chute to the surface of the conveyor belt, resulting in mutual extrusion, stacking and irregular arrangement, thereby preventing automatic, intelligent and high-precision granule detection. The granular crop granule orderly arrangement device can automatically and orderly arrange, spread and expose the visual features of the detected target in the image within the imaging field of view of the image acquisition module, increase the speed and accuracy of image segmentation, identification and classification, improve the identification accuracy of imperfect grains, weeds, insects and impurities in the raw grain, and realize automatic, fast and accurate granule detection and classification of granular crops using machine vision.

[0004] The present application is implemented by the following technical solutions:

[0005] The present application relates to a granular crop grain-by-grain orderly arrangement device, comprising: sequentially arranged orderly arrangement mechanism, vibrating feeding mechanism, material conveying mechanism and image acquisition module, and control system connected with vibrating feeding mechanism and image acquisition module respectively, wherein: the image acquisition module is arranged directly above the material conveying mechanism and outputs the grain arrangement image to the control system, the control system carries out segmentation and detection on the image through machine vision algorithm, thereby detecting the appearance quality of the granular crop and identifying the weeds, insects and impurities contained therein, and the vibrating feeding mechanism receives the vibration instruction from the control system and realizes the orderly arrangement of the granular material through vibration.

[0006] The orderly arrangement mechanism comprises: sequentially connected feed bin and a plurality of groups of horizontally arranged chute, inclined chute and grid teeth, wherein: the lower part of the horizontal chute is provided with a pull-out dust collection box.

[0007] Technical effects

[0008] Compared with the prior art, the present application can continuously, orderly and uniformly arrange the granular crops containing weeds, insects and impurities, such as wheat, soybean, corn and other food crops, or peanuts, rapeseed, coffee beans and other economic crops, in the imaging field of view of the image acquisition module. The raw grain particles falling from different chute channels always maintain a preset interval in the horizontal and vertical directions and assume a posture with the major axis approximately horizontally forward, and in addition, there is no stacking and extrusion between raw grain particles of different sizes. The detection and recognition accuracy of imperfect grains, weeds and insects in the raw grain particles can reach 98% and above, and the quarantine weeds and insects can reach the biological classification of "species".

[0009] The present application realizes the mechanized and automated conveying of granular crops through the vibrating feeding mechanism and the material conveying mechanism; the orderly arrangement mechanism not only uniformly disperses the granular crops, but also avoids stacking and extrusion caused by random motion, so that the granular crops are automatically and orderly arranged and laid in a preset interval in the horizontal and vertical directions; the image acquisition module acquires the images of the regularly arranged crops in the field of view, the control system realizes automatic, rapid and accurate grain-by-grain segmentation, detection and classification of the sample images through the pre-deployed multi-task instance detection and segmentation and fine-grained classification algorithm, greatly reducing the cost of manual detection, thereby improving the detection efficiency and accuracy of imperfect grains, weeds, insects and impurities in the granular crops at the port and in the granary. BRIEF DESCRIPTION OF DRAWINGS

[0010] Figure 1 It is a structural schematic diagram of the present application;

[0011] Figure 2(a) and (b) are the overall schematic diagram and the cross-sectional schematic diagram of the horizontal chute of the orderly arrangement mechanism in the present application (taking wheat as an example);

[0012] Figure 3 (a), (b) and (c) are the isometric view, the sectional view and the cross-sectional schematic diagram of the inclined chute in the present application (taking wheat as an example);

[0013] Figure 4 The flow chart of the quality and quarantine detection algorithm of the granular crop (taking wheat as an example) involved in the present application;

[0014] Figure 5 The working schematic diagram of the quality detection and the weed, insect and impurity detection of 1 kg of wheat in the present application;

[0015] Figure 6 The effect schematic diagram of the embodiment;

[0016] In the figure: 1 orderly arrangement mechanism, 2 vibration feeding mechanism, 21 vibration controller, 22 spring vibrator, 23 feeding hopper, 3 material conveying mechanism, 31 conveying belt, 4 image acquisition module, 41 light source, 42 industrial camera, 5 control system, 6 feeding bin, 7 horizontal chute, 71 dust removal hole, 8 inclined chute, 9 grid tooth, 10 grid, 11 dust collection box, 12 recycling container, H1 minimum height of the feeding hopper from the bottom surface of the feeding bin, L1 length of the horizontal chute, W1 width of the horizontal (inclined) chute, H2 maximum depth of the horizontal chute, θ inclination angle of the inclined chute, L2 length of the grid tooth, L3 length of the inclined chute, W2 minimum width of the grid, H3 maximum height of the grid tooth, AA' cross section of the horizontal chute, BB' sectional view of the inclined chute. DETAILED DESCRIPTION

[0017] As shown in Figure 1 , the present embodiment relates to a granular crop orderly arrangement device, which comprises: orderly arrangement mechanism 1, vibration feeding mechanism 2, material conveying mechanism 3 and image acquisition module 4 arranged in sequence, and control system 5 connected with vibration feeding mechanism 2 and image acquisition module 4 respectively, wherein: image acquisition module 4 is arranged directly above material conveying mechanism 3 and outputs the granular arrangement image to control system 5, control system 5 performs segmentation and detection on the image through machine vision algorithm, thereby performing appearance quality detection on the granular crop and identifying the weeds, insects and impurities contained therein, and vibration feeding mechanism 2 receives instructions from control system 5 to apply linear vibration with corresponding strength and frequency.

[0018] The vibration feeding mechanism 2 comprises a vibration controller 21, a spring oscillator 22 and a feeding hopper 23, wherein the spring oscillator 22 is arranged at the bottom of the beginning end of the ordered arrangement mechanism 1, the feeding hopper 23 is arranged in the inner wall of the front end of the ordered arrangement mechanism 1, and the vibration controller 21 is connected with the control system 5 and the spring oscillator 22 respectively, and outputs vibration with specific amplitude and frequency according to the control instruction, so that the granular crop raw grains slowly fall into the ordered arrangement mechanism 1 from the feeding hopper 23 and are transported forward.

[0019] For 1 kg of wheat raw grains in the embodiment, the volume of the feeding hopper 23 is about 1.5 L, the outlet thereof is a cylinder with an inner diameter of about 12 mm, and the minimum height from the bottom surface of the ordered arrangement mechanism 1 is set to be 10 mm, so as to control the falling speed and flux of the wheat raw grains.

[0020] The material conveying mechanism 3 comprises a conveying belt 31 with a motor arranged at the end of the ordered arrangement mechanism 1, the surface of the conveying belt 31 is attached to the end surface of the ordered arrangement mechanism 1, and the sliding granular crops are automatically and orderly laid into multiple rows with a preset interval and continuously conveyed into the field of view of the image acquisition module 4 through uniform speed transmission.

[0021] For the wheat raw grains in the embodiment, the speed of the material conveying mechanism 3 is set to be 2-4 m / min.

[0022] The image acquisition module 4 comprises a surrounding light source 41 and an industrial camera 42, wherein the industrial camera 42 is arranged directly above the material conveying mechanism 3, and is used for acquiring the image of the granular crops laid in the ordered arrangement on the surface of the conveying belt 31.

[0023] As shown in Figure 2 As shown in (a), the ordered arrangement mechanism 1 comprises a feeding bin 6 and a plurality of groups of horizontally arranged slide grooves 7, inclined slide grooves 8 and grid teeth 9 connected in sequence, wherein the horizontally arranged slide grooves 7 are provided with a pull-out dust collection box 11 below.

[0024] The ordered arrangement mechanism 1 is preferably integrally made of stainless steel with a thickness of about 3 mm.

[0025] The length of the horizontally arranged slide groove 7 is L1, the front end thereof is flush with the end of the feeding bin 6, then the depth thereof gradually increases from the front end to 1 / 2 of the length L1, and then extends to the inclined slide groove 8 with the maximum depth H2, and a plurality of circular dust removal holes 71 with a diameter of about 1 mm are arranged on the bottom of each channel horizontally arranged slide groove 7 from the front end to the end.

[0026] As shown in Figure 2(b) as shown, the horizontal chute 7 is a continuous wavy structure along AA' cross section, the cross section of a single horizontal chute 7 is generally triangular or circular arc shape with two sides of the edge having inclination, and the depth gradually increases, which is to make the wheat grains evenly distributed in different horizontal chutes 7 and orderly arranged in multiple columns while avoiding clogging in a single horizontal chute 7. The width W1, the maximum depth H2 and the length L1 of a single horizontal chute 7 are set according to the type, quantity and apparent size of the granular crops, which is to make the linearly conveyed crop particles evenly and orderly arranged in multiple columns after a sufficient travel in the channel of different horizontal chutes 7 before entering the inclined chute 8.

[0027] In this embodiment, 1kg of wheat is taken as an example, the cross-sectional shape of a single horizontal chute 7 is triangular, and each side is provided with a 1mm round corner to further improve the smoothness, W1 is slightly larger than 2 times the short axis diameter of the wheat grains and is set to 10.4mm, H2 is greater than the height of the wheat grains and is set to 4.6mm, and L1 is set to 350mm.

[0028] As shown in Figure 3 As shown in (a), the inclined chute 8 extends at a preset angle of inclination from the horizontal chute 7, the prismatic teeth 9 and the inclined chute 8 extend upward at the same angle and are regularly arranged at the end of the outlet of the inclined chute 8, so that the teeth 9 are parallel to the horizontal chute 7, and the outlet of each inclined chute 8 is provided with a chamfer of about 60° and 0.7mm in length to flush the bottom surface of the teeth 9. The interval formed by every two teeth 9 constitutes a grid 10 located directly below each inclined chute 8. After the wheat grains fall into the grid 10 from the inclined chute 8, the random displacement due to inertial motion can be greatly eliminated under the restriction of the two tapered teeth 9, so that all the wheat grain particles can be arranged in the same posture (such as the long axis horizontally forward) and laid flat, and the individual large wheat grain particles will not be stagnant between the two teeth 9, and finally the uniform driving of the material conveying mechanism 3 makes each column of wheat grain particles automatically arranged into multiple rows in the same interval.

[0029] As shown in Figure 3 (b) is Figure 3 (a) is a cross-sectional view of the inclined chute 8 along BB', the angle of inclination between the inclined chute 8 and the horizontal chute 7 is θ, the teeth 9 extend forward at the same angle along the end of the inclined chute 8, and the length L2 is generally set to 25mm according to the size of 1kg of wheat grains used in this embodiment. The angle of inclination θ and the length L2 of the inclined chute 8 are determined according to the size of the wheat grains. The angle of inclination θ and the length L2 are determined by theoretical calculation under certain conditions.

[0030] Preferably, the angle of inclination θ is Wherein: H is the vertical height from the horizontal chute 7 to the surface of the conveying belt, μ is the surface friction coefficient of the inclined chute 8, The length of the inclined chute 8. Ideally, the speed of the wheat kernels remains unchanged before and after sliding in the inclined chute 8, and the relative speed of the conveyor belt 31 pulls apart the front and back spacing when the kernels slide onto the surface of the conveyor belt 31. The inclined chute 8 is designed to have an appropriate inclination angle θ, so that after the kernels are uniformly and orderly arranged into multiple columns in the inclined chute 8, they can smoothly slide onto the surface of the conveyor belt 31 under the restriction of the side walls of the grid teeth 9 and be automatically arranged and tiled into multiple rows at a predetermined spacing by the relative speed between the material conveying mechanism 3, without accumulating at the junction of the grid teeth 9 and the conveyor belt 31, and avoiding the kernels being too sparse due to the excessive speed of the material conveying mechanism 3, which affects the acquisition flux of the image acquisition module 4.

[0031] In this embodiment, the vibration frequency of the vibrating feeding mechanism 2 is 40-45 Hz, the height of the horizontal chute 7 from the surface of the conveyor belt 31 is 30-50 mm, the driving speed of the material conveying mechanism 3 is 2-4 m / min, the average width of the wheat kernels is 4.5 mm, and the rolling friction coefficient of the surface of the stainless steel inclined chute 8 is about 0.3. The minimum value of the inclination angle θ is calculated to be 13.5°, and the length of the inclined chute 8 is 120 mm. Considering the actual application scenarios and engineering errors, the θ in this embodiment is set to 15°.

[0032] As shown in Figure 3 (c), the grid teeth 9 adopt a pyramidal structure that decreases radially or a semi-circular conical structure. The maximum height H3 near the outlet side of the inclined chute 8 is usually similar to the thickness of the granular crops. The grid 10 formed by the equidistant distribution of the grid teeth 9 is located directly below the corresponding inclined chute 8, and its width gradually increases. The minimum width W2 is usually equal to the width of the crop particles.

[0033] In this embodiment, the grid teeth 9 adopt a three-pyramidal structure, and the maximum height H3 and the minimum width W2 of the grid 10 are set to 4.5 mm and 3.5 mm, respectively.

[0034] As shown in Figure 4 , the machine vision algorithm includes:

[0035] 1) After the image acquisition module 4 acquires the image of the orderly arranged and tiled wheat kernels, the control system 5 pre-processes the original image.

[0036] The pre-processing includes cropping, linear transformation, filter denoising, and distortion correction.

[0037] 2) A Mask R-CNN model is used for multi-task instance detection and segmentation, which simultaneously obtains the coarse classification and instance of each kernel in the image.

[0038] ​The Mask R-CNN model is trained by using a wheat, weed and insect detection and instance segmentation strong supervision data set.

[0039] The wheat, weed and insect detection and instance segmentation strong supervision data set is constructed by shooting, collecting from the network and providing pictures of wheat raw grains, various weeds and insects by grain storage institutions and performing relevant labeling.

[0040] The coarse classification includes crops, weeds, insects and impurities.

[0041] The instance includes each segmentation instance of crops, each segmentation instance of weeds, each segmentation instance of insects and each segmentation instance of impurities.

[0042] When the instance obtained in step 2) does not belong to wheat and the existing weeds and insects in the data set, the instance is regarded as other impurities for quantity statistics and labeling.

[0043] 3) The instances of coarse classification of wheat, weeds and insects are respectively sent to corresponding fine classification models, and the quality of wheat is detected, the category and quantity of imperfect grains are labeled, and the species classification and quarantine labeling of weeds and insects are performed.

[0044] The corresponding fine classification model is trained by using a wheat quality, weed and insect weak supervision fine-grained classification data set.

[0045] The wheat quality, weed and insect weak supervision fine-grained classification data set is constructed by shooting, collecting from the network and providing pictures of perfect grains, insect-eaten grains, germinated grains and moldy grains and the like of wheat, and the pictures of weeds and insects of different families, genera and species and performing relevant labeling.

[0046] The corresponding fine classification model refers to an imperfect grain fine classification model, a weed fine classification model and an insect fine classification model.

[0047] The imperfect grain includes insect-eaten grains, germinated grains and moldy grains and the like.

[0048] 4) After completing all image acquisition, the results obtained in step 3) are counted and displayed, and the quality information of the batch of wheat raw grains, the species quantity and quarantine of weeds and insects are displayed.

[0049] As Figure 5As shown, through specific actual experiments, the appearance quality and weeds, insects and impurities of 1 kg of wheat raw grains were precisely detected, specifically including: setting the amplitude of the vibration controller 21 to 123 V and the frequency to 42 Hz through the control system 5, and starting the spring vibrator 22 to transmit linear vibration to the orderly arrangement mechanism 1 and the feeding hopper 23 arranged inside, and setting the transmission speed of the material conveying mechanism 3, so that the surface of the conveying belt 41 is driven at a constant speed to match the bottom surface of the end of the orderly arrangement mechanism 1. 1 kg of wheat raw grains was poured into the feeding hopper 23, under the driving of gravity and linear vibration, the raw grains fell from the columnar outlet at the bottom of the feeding hopper 23 into the feed bin 6 at the front end of the orderly arrangement mechanism 1. The proper height of the outlet of the feeding hopper 23 from the bottom surface of the feed bin 6 not only ensures the uniform falling of the wheat raw grains, but also produces a preset flow limiting effect to avoid large accumulation. The falling wheat raw grains continuously move on the slightly inclined bottom surface of the feed bin 6 and gradually disperse into the horizontal chute 7. The horizontal chute 7 with gradually increasing depth not only enables the granular wheat raw grains to be uniformly distributed to different channels to avoid accumulation at the front end, but also enables them to be automatically, orderly and regularly arranged into multiple columns, and the dust removal holes 71 at the bottom of each channel effectively screen out dust and other small impurities in the wheat raw grains, and the lower pull-out dust collection box 11 achieves the purpose of regular cleaning. When the wheat raw grain particles are orderly arranged in multiple columns in the horizontal chute 7, some broken particles or small-scale weeds, insects and other detection objects still exist in the form of stacking or adhesion, and after the raw grains enter the inclined chute 8, the above-mentioned small-scale detection objects will be further laid flat and form a front-to-back orderly arrangement pattern in the channels of the inclined chute 8. The chamfer at the outlet of the inclined chute 8 eliminates the random movement generated when the wheat raw grains slide out, so that they fall smoothly into the grid 10 formed by the grid teeth 9 at the end. The tapered grid teeth 9 with gradually decreasing height and the grid 10 with gradually increasing width not only enable the falling wheat raw grain particles to keep a consistent posture (such as long axis forward) and lay flat stably, but also avoid individual large-size particles from being retained on the surface. The relative speed between the material conveying mechanism 3 and the wheat raw grains enables the wheat raw grain particles in each column falling from different channels of the inclined chute 8 to be orderly arranged into multiple rows at a proper distance and laid flat regularly on the surface of the conveying belt 31, and enter the field of view of the image acquisition module 4 through uniform driving. The light sources 41 arranged on four sides provide uniform and stable illumination for the wheat raw grains orderly arranged and laid flat on the surface of the conveying belt 31, the industrial camera 42 continuously acquires images in the field of view and transmits them into the control system 5, and through the multi-task instance detection and segmentation algorithm and the weakly supervised fine-grained classification algorithm deployed in advance, each frame of image is detected, recognized and classified, the types and quantities of imperfect grains, weeds, insects and impurities contained in the batch of wheat raw grains are counted and displayed, and the detected quarantine species are marked. The wheat raw grains after image acquisition continuously move forward and finally fall into the recycling container 12 arranged at the end of the conveying belt 31.

[0050] Through the above process, 1 kg of wheat raw grains can be smoothly, orderly and grain by grain laid on the surface of the conveyor belt 31 and continuously conveyed into the field of view of the image acquisition module 4 within 15 minutes, with an average image number of 420 frames, as shown in Figure 6 To achieve accurate detection and identification of each grain, the field of view of the image acquisition module 4 is 120x90 mm, ensuring that the imaging accuracy is 75 pixels / mm. The average single field of view of the image acquisition module 4 contains about 120 wheat raw grain particles, 8 rows in the longitudinal direction, and the average distance between each row is about 6 mm; each row contains an average of about 10 wheat raw grain particles with the long axis horizontal forward, and the average horizontal distance between each particle is about 1.5 mm. In the above process, the probability of wheat raw grain particles being blocked in the horizontal chute 7, the inclined chute 8 or the grid 10 is less than 0.1%, and the probability of random movement on the surface of the conveyor belt 31 causing mutual extrusion or stacking is less than 0.05%. Through continuous imaging of the image acquisition module 4, the device can segment and identify weeds, insects and impurities contained in 1 kg of wheat raw grains within 40 minutes using multi-task instance detection and segmentation algorithms deployed in the control system 5, and then obtain the attributes of wheat imperfect grains and weeds and insects in biological classification using corresponding weakly supervised fine-grained classification algorithms, which can reach the "species" level at most, and the recognition accuracy is maintained at more than 98%.

[0051] Compared with the prior art, the present application sets a horizontal and inclined chute with gradually increasing depth and continuous wavy cross-sectional shape in the device structure; and a tapered grid tooth with gradually decreasing height at the end of the inclined chute, and a grid with gradually increasing width below the outlet of each chute channel. In the target detection and identification algorithm, the present application combines multi-task instance detection and segmentation algorithm and weakly supervised fine-grained classification algorithm. Compared with the prior art, the vibrating feeding mechanism of the present application sets the vibration frequency, amplitude of the spring vibrator and the external dimensions of the feeding hopper through the vibration controller for different types and quantities of granular crops, thereby providing appropriate power for feeding and conveying. The orderly arrangement mechanism effectively limits the mutual accumulation, extrusion and irregular arrangement of granular crops through the cooperation of the feeding bin, horizontal chute and inclined chute, so that the granular crops are automatically and orderly distributed into multiple columns and smoothly slide onto the surface of the conveyor belt. The bottom surface of the feeding bin is inclined downward at a small angle, so that the crops falling from the feeding hopper are quickly and evenly distributed under the action of vibration and gravity and enter the horizontal chute. The gradually increasing depth of the front half of the horizontal chute enables the granular crops to be evenly distributed into different channels, and the maximum depth of the rear half limits the granular crops from crossing different channels and automatically and orderly arranging into multiple columns before sliding to the inclined chute; the dust removal holes uniformly distributed from the front end to the end of the bottom of each horizontal chute can effectively screen out dust and small impurities in the granular crops and collect and clean them through the dust collection box at the bottom, greatly improving the clarity of the granular crops during imaging. The inclination angle between the inclined chute and the horizontal chute, the extension length of the inclined chute, the height of the horizontal chute from the surface of the conveyor belt and the conveying speed of the material conveying mechanism are matched through theoretical calculation, ensuring that each column of granular crops is neither accumulated nor excessively dispersed when smoothly sliding onto the surface of the conveyor belt through the inclined chute; the inclined chute also enables the small-scale granular crops continuously stacked in the horizontal chute to be further flattened and orderly arranged. The tapered grid tooth and the inclined chute maintain a preset included angle and extend forward while being parallel to the horizontal chute and being uniformly arranged at the end of the inclined chute, thereby not only making the grid tooth fit the surface of the conveyor belt, but also forming a grid with gradually increasing width directly below each inclined chute, so that the granular crops arranged in multiple columns can avoid random movement and smoothly slide onto the surface of the conveyor belt while maintaining a consistent posture and being automatically and orderly flattened between the grids, and in addition, large-sized granular crops are prevented from being trapped between adjacent grid teeth. The transmission speed of the material conveying mechanism matches the sliding speed of the granular crops, so that each column of granular crops in different inclined chute channels is automatically and evenly flattened into multiple rows with a preset spacing between the front and back, and continuously conveyed into the imaging field of the image acquisition module.The image acquisition module acquires and transmits the images of the orderly arranged and tiled granular crops in the field of view in a time manner, and the multi-task instance in the control system can detect and segment the model in combination with the appearance quality detection, weed and insect fine-grained classification model, so that the biological population and quarantine information of imperfect grains, weeds and insects and other impurities in the sample images can be detected and recognized grain by grain, accurately and efficiently. The application improves the efficiency and precision of the detection of the granular crops by using the machine vision technology, is a key core component for the appearance quality and quarantine detection of the granular crops grain by grain, and provides strong technical support for the automatic, rapid and accurate detection of the imported crops at the port and the stored grains.

[0052] The above specific embodiments can be adjusted in different ways by those skilled in the art without departing from the principles and purposes of the application, the protection scope of the application is subject to the claims and is not limited by the above specific embodiments, and each implementation scheme within the scope is subject to the constraints of the application.

Claims

1. A visual detection method based on a device for orderly arrangement of granular crops, characterized in that, The arrangement device includes: an ordered arrangement mechanism, a vibrating feeding mechanism, a material conveying mechanism, and an image acquisition module arranged in sequence, and a control system connected to the vibrating feeding mechanism and the image acquisition module respectively. The image acquisition module is located directly above the material conveying mechanism and outputs the particle arrangement image to the control system. The control system uses machine vision algorithms to segment and detect the image, thereby performing appearance quality detection on the granular crops and identifying weeds, insects, and other impurities contained therein. The vibrating feeding mechanism receives vibration commands from the control system and achieves the ordered arrangement of particles through vibration. The ordered arrangement mechanism includes: feed bins connected in sequence and several sets of horizontal chutes, inclined chutes and grid teeth arranged in parallel, wherein: a pull-out dust collection box is provided below the horizontal chutes; The front end of the horizontal chute is flush with the end of the feed hopper, and its depth gradually increases from the front end to half of its length, and then extends to the inclined chute. Several circular dust removal holes are provided at the bottom of each channel horizontal chute from the front end to the end. The horizontal chute has a continuous wave-like structure in the cross-sectional direction. The cross-section of a single horizontal chute is triangular or arc-shaped and the depth gradually increases. Its width is twice the diameter or width of the granular crop, and the maximum depth is equal to the diameter or thickness of the granular crop. The grid teeth adopt a radially tapered pyramid or semi-conical structure. The maximum height near the outlet of the inclined chute matches the thickness of the granular crop. The grid formed by the equally spaced grid teeth is located directly below the corresponding inclined chute, and its width gradually increases, with the minimum width being equal to the width of the crop particles. The angle between the inclined slide and the horizontal slide is θ, and the grid teeth extend forward along the end of the inclined slide at the same angle; The aforementioned tilt angle Where: H is the vertical height from the horizontal chute to the surface of the conveyor belt, and μ is the coefficient of friction of the inclined chute surface. The length of the inclined chute; The aforementioned visual detection method includes: 1) After receiving the image of the orderly arranged and laid-out raw crops acquired by the image acquisition module, the control system preprocesses the original image; 2) The Mask R-CNN model is used for multi-task instance detection and segmentation, and the coarse classification and instance of each grain of rice in the image are obtained simultaneously; The Mask R-CNN model described above was trained using a strongly supervised dataset for the detection and instance segmentation of crops, weeds, and insects. The strongly supervised dataset for the detection and instance segmentation of crops, weeds and insects was constructed by taking pictures, collecting them from the Internet and providing pictures of raw crops, various types of weeds and insects provided by grain storage institutions, and then annotating them accordingly. The coarse classification includes: crops, weeds, insects, and other impurities; The examples include: each segmented example of crops, each segmented example of weeds, each segmented example of insects, and each segmented example of other impurities; If the instance obtained in step 2) does not belong to crops or weeds and insects already in the dataset, it is counted and labeled as other impurities. 3) The instances that are roughly classified into crops, weeds and insects are fed into the corresponding fine classification models respectively. The quality of crops is tested and the category and quantity of imperfect grains are labeled. The species classification and quarantine labeling of weeds and insects are performed respectively. The aforementioned crop quality, weeds and insects weakly supervised fine-grained classification datasets were constructed by taking pictures, collecting images from the internet, and providing images of perfect and imperfect crop grains, as well as weeds and insects belonging to different families, genera and species, from grain storage institutions, and then annotating them accordingly. The corresponding fine classification models mentioned above refer to: imperfect fine classification model, weed fine classification model, and insect fine classification model; The imperfect particles include: insect-eaten particles, sprouted particles, and moldy particles; 4) After all images have been acquired, the results obtained in step 3) are statistically analyzed and displayed to show the quality information, weed and insect species, and quarantine status of the batch of raw agricultural products.

2. The visual detection method according to claim 1, characterized in that, The vibrating feeding mechanism includes a vibration controller, a spring vibrator, and a feeding hopper. The spring vibrator is located at the bottom of the beginning of the orderly arrangement mechanism, and the feeding hopper is located in the inner wall of the front end of the orderly arrangement mechanism. The vibration controller is connected to the control system and the spring vibrator respectively. According to the control command, it outputs vibration with a specific amplitude and frequency, so that the granular agricultural raw materials fall slowly from the feeding hopper into the orderly arrangement mechanism and are conveyed forward.

3. The visual detection method according to claim 1, characterized in that, The material conveying mechanism includes a motor-driven conveyor belt located at the end of the orderly arrangement mechanism. The surface of the conveyor belt is in contact with the end surface of the orderly arrangement mechanism. Through uniform speed transmission, each row of granular crops that has fallen down is automatically and orderly laid out into multiple rows at a preset interval and continuously conveyed into the field of view of the image acquisition module.

4. The visual detection method according to claim 1, characterized in that, The image acquisition module includes a surround light source and an industrial camera, wherein the industrial camera is positioned directly above the material conveying mechanism and is used to acquire images of granular crops that are regularly arranged and laid flat on the surface of the conveyor belt.

Images