Vegetable seed sowing device based on machine vision and detection method

Abstract

The invention discloses a vegetable seed sowing device based on machine vision and a detection method, and relates to the technical field of sowing devices. The two rows of turning plate mechanisms are arranged in parallel and are arranged in the rack, each row of turning plate mechanism comprises twelve independent turning plates, the tops of the turning plates are provided with hole trays, the bottoms of the turning plates are hinged to two air cylinders, one air cylinder drives the turning plates to turn over towards one side, and the other air cylinder drives the turning plates to turn over towards the other side; the industrial visual inspection assembly is arranged in the rack, corresponds to the two rows of turning plates, and is used for collecting seed images in each hole tray in real time and identifying the number of seeds; the discharging hopper is arranged between the two rows of turning plates; the two return hoppers are respectively arranged on the outer sides of the two rows of turning plates; the control system is arranged in the rack and is used for controlling the industrial visual detection assembly and the air cylinder, so that efficient and rapid seeding is realized, double-row complementary logic is combined, the miss-seeding rate and the multi-seeding rate are reduced, the quantity of seeds in each hole conforms to a set value, and the error is low.

Description

Technical Field

[0001] The present invention relates to the technical field of sowing devices, and more specifically, to a vegetable seed sowing device and detection method based on machine vision. Background Art

[0002] Traditional vegetable seed sowing devices mostly adopt mechanical sowing structures or semi-automatic equipment, which have prominent problems such as low sowing accuracy, inability to sort defective seeds in real time, and poor equipment adaptability. Specifically, mechanical seeders rely on fixed hole positions or vibration sorting, making it difficult to accurately control the sowing amount per hole, often resulting in empty holes or over-sowing, leading to waste of seeds.

[0003] Based on this, the present invention provides a vegetable seed sowing device and detection method based on machine vision. Summary of the Invention

[0004] In order to solve the problems raised in the above background art, the present invention provides a vegetable seed sowing device and detection method based on machine vision.

[0005] The vegetable seed sowing device based on machine vision provided by the present invention adopts the following technical solutions:

[0006] A vegetable seed sowing device based on machine vision includes a frame; two rows of flap mechanisms arranged in parallel, which are arranged inside the frame. Each row of flap mechanisms includes twelve independent flaps, and a seed tray is arranged on the top of the flap for accommodating seeds. Two cylinders are hinged at the bottom of the flap. One cylinder drives the flap to flip to one side for sowing qualified seeds, and the other cylinder drives the flap to flip to the other side for discarding unqualified seeds; an industrial vision detection component is arranged inside the frame and corresponds to the two rows of flaps for real-time collecting seed images in each seed tray and identifying the number of seeds; a feeding hopper is arranged between the two rows of flaps for receiving qualified seeds for sowing; two return hoppers are respectively arranged outside the two rows of flaps for recycling the unqualified seeds discarded by the flaps; a control system is arranged inside the frame for controlling the industrial vision detection component and the cylinders.

[0007] Preferably, the industrial vision detection component includes a high-resolution camera and an image processing unit. The camera synchronously takes pictures of the seed tray images of the two rows of flaps from a top-down perspective, and the image processing unit counts the number of seeds through an edge detection algorithm.

[0008] Preferably, the edge detection algorithm adopts an improved gradient amplitude calculation model, and the gradient operator is defined as:

[0009]

[0010] Where:

[0011] ω x =1+α·var(I(x-1:x+1,y)),ω y =1+α·var(I(x,y-1:y+1)),

[0012] α is the local texture sensitivity coefficient, var represents the grayscale variance of the neighborhood pixels, and the edge features of the high texture area are enhanced through dynamic weights.

[0013] The present invention provides a method for detecting a vegetable seed sowing device based on machine vision, which adopts the following technical solution:

[0014] A detection method for a vegetable seed sowing device based on machine vision, based on the above-mentioned vegetable seed sowing device based on machine vision, comprises the following steps:

[0015] S1, synchronously adsorbing seeds on the plug trays of two rows of flaps;

[0016] S2, identifying the number of seeds in each plug tray through an industrial visual inspection component and generating inspection data;

[0017] S3. The control system compares the detection data with the preset threshold and performs the following operations:

[0018] If at least one of the two rows of plug trays at the same longitudinal position meets the threshold, the qualified flip-over seeding is triggered and the position of the unqualified plug tray is recorded;

[0019] If neither meets the threshold, two flaps are triggered to discard the seed;

[0020] S4. In the next sowing cycle, the adsorbed seeds are added to the recorded hole trays that need to be reseeded first.

[0021] Preferably, in S2, the seed quantity recognition adopts a multi-scale morphological reconstruction algorithm, and the positioning seed region segmentation function is:

[0022]

[0023] in:

[0024] B k is the k-scale structural element, Represents morphological expansion, and eliminates particles with a size smaller than the threshold d through iteration. min The noise particles are removed and the effective seed outline is retained.

[0025] Preferably, in S3, the preset threshold is an adaptive range threshold, and the calculation formula is:

[0026]

[0027] in:

[0028] A cell is the area of the plug tray unit, r seed is the nominal radius of the seed, Δr is the particle size tolerance, β∈[0.8, 1.2] is the seeding density coefficient, and Δr is adjusted in real time through the pressure sensor at the end of the robotic arm.

[0029] Preferably, in S4, the reseeding priority calculation adopts a Markov decision model, and the state transition probability is defined as:

[0030]

[0031] Where:

[0032] s=(xmiss, tdela y , nretry) represents the state vector of the missing position, delay time and number of repetitions, a∈{reseed immediately, delay reseeding} is the action space, and the optimal reseeding strategy is updated online through the Q-learning algorithm.

[0033] Preferably, anisotropic diffusion preprocessing is performed before morphological reconstruction:

[0034]

[0035] Where:

[0036] K = 0.3, is the thermal conductivity, the diffusion intensity along the long axis direction of the seed is increased by 50%, and the geometric integrity of the flat seed is effectively maintained.

[0037] Preferably, a U-Net segmentation network is integrated after the diffusion treatment, and the loss function is defined as:

[0038]

[0039] Where:

[0040] α ＝ 0.3 is the balance coefficient, BCE is the binary cross entropy, 3 times the weight is assigned to the seed edge pixels during the Dice coefficient calculation, and the intersection over union of the adhered seeds segmentation reaches 92.7%.

[0041] In summary, the present invention includes the following beneficial technical effects:

[0042] 1. After the plug trays are synchronously adsorbed with seeds by two rows of flap mechanisms, the high-resolution cameras of the industrial vision detection components synchronously take seed images of the two rows of plug trays from a top-down perspective. Through the image processing unit combined with the edge detection algorithm, the number of seeds in each plug tray is statistically calculated in real time. Then, the control system compares the detection result with a preset value (such as 1 seed per hole). If one of the two rows of plug trays at the same position (such as No. 1 and No. A) is qualified and the other is unqualified, the complementary mechanism is triggered, and only the qualified seeds are retained for sowing, while the unqualified seeds are removed; if both rows are qualified, single-row sowing is default selected to avoid duplication; if both are unqualified, it is marked for reseeding. According to the judgment result, the cylinder corresponding to the qualified seeds drives the flap to turn inward, and the seeds are put into the hopper to complete sowing; the unqualified seeds are driven by another cylinder to turn outward and fall into the return hopper for recycling. The system records the positions of the unqualified holes, and the missing seeds are preferentially supplemented during the next seed suction. The cycle of detection and sorting is carried out until all holes meet the standards, ensuring the integrity of sowing. Through this design, efficient and rapid sowing is achieved. Combining the double-row complementary logic, the rates of missed sowing and over-sowing are reduced, and the number of seeds per hole strictly conforms to the set value with low error.

[0043] 2. Based on the multi-scale morphological reconstruction algorithm, iterative dilation operations are performed on the seed images using structural elements of different scales to dynamically remove noise interferences such as dust and debris with particle sizes smaller than the threshold, while retaining the complete seed contours (especially for seeds with large particle size differences or slight adhesions), resulting in improved segmentation accuracy and a lower misjudgment rate compared to traditional threshold segmentation methods. And the algorithm adaptively matches the particle size distribution ranges (0.5 - 5 mm) of vegetable seeds such as tomatoes and cabbages by defining a scale-sensitive segmentation function, effectively solving the problems of missed detection of small seeds and over-segmentation of large seeds in a high-noise environment; at the same time, combined with the high-resolution imaging of the industrial camera, a single detection can synchronously process the rapid analysis of two rows of 24-hole seeds, effectively shortening the detection cycle.

[0044] 3. Based on the anisotropic diffusion model (thermal conductivity K = 0.3), the diffusion intensity is enhanced by 50% along the long axis direction of the seeds to specifically retain the geometric features (such as aspect ratio, edge curvature) of flat seeds such as cucumbers and pumpkins, avoiding the detail blurring caused by traditional Gaussian filtering. By adjusting the diffusion direction weights (such as switching to isotropic diffusion for round seeds), the preprocessing requirements of seeds with different geometric characteristics such as tomatoes and cabbages can be adapted.

[0045] The above summary is only for the purpose of the specification and is not intended to be limiting in any way. In addition to the illustrative aspects, embodiments, and features described above, further aspects, embodiments, and features of the present invention will become readily apparent by reference to the drawings and the following detailed description. Description of the Drawings

[0046] Figure 1It is a schematic structural diagram of a vegetable seed sowing device based on machine vision in an embodiment of the present invention;

[0047] Figure 2 It is a schematic internal structural diagram of a vegetable seed sowing device based on machine vision in an embodiment of the present invention;

[0048] Figure 3 It is a schematic diagram of the other side inside a vegetable seed sowing device based on machine vision in an embodiment of the present invention;

[0049] Figure 4 It is a schematic sectional structural diagram of a vegetable seed sowing device based on machine vision in an embodiment of the present invention.

[0050] Explanation of reference numerals: 1, frame; 2, flap; 3, seedling tray; 4, cylinder; 5, industrial vision detection component; 6, hopper; 7, return hopper. Detailed implementation manners

[0051] The following further elaborates on the present invention in conjunction with the attached Figures 1 to 4 for a more detailed description.

[0052] It should be noted that the drawings are schematic and not drawn to scale. For clarity and convenience in the figures, the relative sizes and proportions of the parts shown in the figures are exaggerated or reduced in size for illustration, and any dimensions are merely exemplary and not limiting. Additionally, the same reference numerals are used for the same structures, elements, or fittings that appear in more than two figures to represent similar features.

[0053] Embodiment 1

[0054] An embodiment of the present invention discloses a vegetable seed sowing device based on machine vision. Referring to Figures 1 to 4 , a vegetable seed sowing device based on machine vision includes a frame 1; two rows of parallel flap mechanisms arranged inside the frame 1, each row of flap mechanisms including twelve independent flaps 2, and a seedling tray 3 is arranged on top of the flap 2 for accommodating seeds. Two cylinders 4 are hinged at the bottom of the flap 2. One cylinder 4 drives the flap 2 to flip to one side for sowing qualified seeds, and the other cylinder 4 drives the flap 2 to flip to the other side for discarding unqualified seeds; an industrial vision detection component 5 is arranged inside the frame 1 and corresponds to the two rows of flaps 2 for real-time collecting images of seeds in each seedling tray 3 and identifying the number of seeds; a hopper 6 is arranged between the two rows of flaps 2 for receiving qualified seeds for sowing; two return hoppers 7 are respectively arranged outside the two rows of flaps 2 for recycling the unqualified seeds discarded by the flaps 2; a control system is arranged inside the frame 1 for controlling the industrial vision detection component 5 and the cylinders 4.

[0055] Specifically, the industrial vision inspection component 5 includes a high-resolution camera and an image processing unit. The camera synchronously captures images of the seed trays 3 of the two rows of flap mechanisms 2 from a top-down perspective, and the image processing unit counts the number of seeds through an edge detection algorithm.

[0056] Specifically, after the seed trays 3 of the two rows of flap mechanisms synchronously adsorb seeds, the high-resolution camera of the industrial vision inspection component 5 synchronously captures images of the seeds in the two rows of seed trays 3 from a top-down perspective. Through the image processing unit combined with the edge detection algorithm, the number of seeds in each seed tray 3 is counted in real time. Then, the control system compares the detection result with a preset value (such as 1 seed per hole). If one of the two rows of seed trays 3 at the same position (such as No. 1 and No. A) is qualified and the other is unqualified, the complementary mechanism is triggered, and only the qualified seeds are retained for sowing, while the unqualified seeds are removed; if both rows are qualified, single-row sowing is default selected to avoid duplication; if both are unqualified, it is marked for reseeding. According to the judgment result, the cylinder 4 corresponding to the qualified seeds drives the flap 2 to flip inward, and the seeds are put into the feeding hopper 6 to complete sowing; the unqualified seeds are driven by another cylinder 4 to flip outward and fall into the return hopper 7 for recycling. The system records the positions of the unqualified holes, and the missing seeds are preferentially supplemented during the next seed suction. The cycle of detection and sorting continues until all holes meet the standard, ensuring the integrity of sowing. Through this design, efficient and rapid sowing is achieved. Combining the double-row complementary logic, the rates of missed sowing and over-sowing are reduced, and the number of seeds per hole strictly conforms to the set value with low error.

[0057] Specifically, the edge detection algorithm adopts an improved gradient magnitude calculation model, and the gradient operator is defined as:

[0058]

[0059] where:

[0060] ω x = 1 + α·var(I(x - 1:x + 1, y)), ω y = 1 + α·var(I(x, y - 1:y + 1)),

[0061] α is the local texture sensitivity coefficient, and var represents the gray variance of neighboring pixels, enhancing the edge features of high-texture regions through dynamic weights.

[0062] The improved gradient magnitude calculation model enhances the edge feature extraction ability of high-texture regions through dynamic weights (based on the local texture sensitivity coefficient and gray variance), improves the counting accuracy of dense and sticky seeds, and reduces the missed detection rate.

[0063] Specifically, the cylinder 4 adopts double-pressure fuzzy control, and the cooperative control parameters are defined as:

[0064]

[0065] Wherein:

[0066] K p = μ·e -λt is a time-varying proportionality coefficient, and K d is the differential term of the pressure difference. The control parameters μ and λ are optimized through the Lyapunov stability criterion.

[0067] The cylinder 4 adopts a double-pressure fuzzy control strategy. Through the collaborative optimization of the time-varying proportionality coefficient and the differential term of the pressure difference, combined with the Lyapunov stability criterion to dynamically adjust the parameters, it ensures the rapid response and low jitter error of the sorting action of the flap 2, and avoids seed spillage.

[0068] Specifically, the image processing unit integrates a Siamese neural network, and the feature similarity metric is defined as:

[0069]

[0070] Wherein:

[0071] F i , F j are the feature vectors of different seed trays 3, H(·) is the color histogram feature, and γ is a learnable parameter. Multi-variety seed classification is achieved through contrast learning.

[0072] Combined with the contrast learning mechanism of multiple features (color histogram, learnable shape parameter) of the Siamese neural network, rapid classification and abnormal rejection of multi-variety seeds are realized, the classification accuracy is improved, and accurate recognition of seeds with complex shapes such as tomatoes and peppers is adapted.

[0073] By deeply integrating edge computing, stability control and deep learning, the accuracy and efficiency bottlenecks of traditional sowing equipment in the case of complex seed shapes and pipeline environments are solved.

[0074] Embodiment 2

[0075] A detection method for a vegetable seed sowing device based on machine vision, based on the above-mentioned vegetable seed sowing device based on machine vision, includes the following steps:

[0076] S1. Synchronously adsorb seeds on the seed trays 3 of the two rows of flaps 2;

[0077] S2. Identify the number of seeds in each seed tray 3 through the industrial vision detection component 5 to generate detection data;

[0078] S3. The control system compares the detection data with a preset threshold and performs the following operations:

[0079] If at least one of the two rows of seed trays 3 at the same longitudinal position meets the threshold, trigger the qualified flap 2 to sow and record the position of the unqualified seed tray 3;

[0080] If none of them meets the threshold, trigger the two flap gates 2 to discard the seeds;

[0081] S4. In the next sowing cycle, preferentially supplement the adsorbed seeds to the recorded plug trays 3 that need to be reseeded.

[0082] Specifically, in S2, the seed quantity recognition adopts a multi-scale morphological reconstruction algorithm, and the positioning seed region segmentation function is:

[0083]

[0084] Where:

[0085] B k is the k-scale structural element, represents morphological dilation, and the noise particles with a particle size smaller than the threshold d are eliminated through iteration min to retain the effective seed contour.

[0086] Based on the multi-scale morphological reconstruction algorithm, the seed image is iteratively dilated with structural elements of different scales to dynamically eliminate noise interferences such as dust and debris with a particle size smaller than the threshold, while retaining the complete seed contour (especially for seeds with large particle size differences or slight adhesion), improving the segmentation accuracy, reducing the misjudgment rate compared with the traditional threshold segmentation method, and the algorithm adaptively matches the particle size distribution range (0.5 - 5 mm) of vegetable seeds such as tomatoes and cabbages by defining a scale-sensitive segmentation function, effectively solving the problems of missed detection of small seeds and over-segmentation of large seeds in a high-noise environment; at the same time, combined with the high-resolution imaging of the industrial camera, the rapid analysis of two rows of 24-hole seeds can be synchronously processed in a single detection, effectively shortening the detection cycle.

[0087] Specifically, in S3, the preset threshold is an adaptive range threshold, and the calculation formula is:

[0088]

[0089] Where:

[0090] A cell is the unit area of the plug tray (3), r seed is the nominal radius of the seed, Δr is the particle size tolerance, and β ∈ [0.8, 1.2] is the seeding density coefficient, and Δr is adjusted in real time through the pressure sensor at the end of the robotic arm.

[0091] Based on the nominal radius of the seed, the particle size tolerance and the seeding density coefficient, combined with the unit area of the plug tray 3, an adaptive decision threshold model is constructed to effectively solve the misjudgment problem caused by the seed size difference (such as the particle size of cucumber seeds and celery seeds differs by 5 times) of the traditional fixed threshold.

[0092] Specifically, in S4, the reseeding priority calculation adopts a Markov decision model, and the state transition probability is defined as:

[0093]

[0094] Where:

[0095] s = (xmiss, tdela y , nretry) represents the state vector of the missing position, delay time, and number of repetitions. a ∈ {reseed immediately, reseed with delay} is the action space, and the optimal reseeding strategy is updated online through the Q-learning algorithm.

[0096] Based on the state transition probability of the Markov model (covering multi-dimensional state vectors such as missing position, delay time, and number of repetitions), the optimal reseeding action strategy is iteratively updated online through the Q-learning algorithm, dynamically weighing the reseeding efficiency and resource consumption. For example, preferentially processing high-density missing areas or acupoints with historical reseeding failures can improve the reseeding success rate.

[0097] Specifically, anisotropic diffusion preprocessing is performed before morphological reconstruction:

[0098]

[0099] Where:

[0100] K = 0.3, is the thermal conductivity, and the diffusion intensity is increased by 50% along the long axis direction of the seed, effectively maintaining the geometric integrity of flat seeds.

[0101] Based on the anisotropic diffusion model (thermal conductivity K = 0.3), the diffusion intensity is enhanced by 50% along the long axis direction of the seed, specifically retaining the geometric features (such as aspect ratio and edge curvature) of flat seeds such as cucumbers and pumpkins, avoiding the detail blurring caused by traditional Gaussian filtering, and by adjusting the diffusion direction weight (such as switching to isotropic diffusion for round seeds), the preprocessing requirements of seeds with different geometric characteristics such as tomatoes and cabbages can be adapted.

[0102] Specifically, after the diffusion process, the U-Net segmentation network is integrated, and the loss function is defined as:

[0103]

[0104] Where:

[0105] α ＝ 0.3 is the balance coefficient, BCE is the binary cross-entropy, and 3 times the weight is assigned to the seed edge pixels during the Dice coefficient calculation, achieving a 92.7% intersection over union for the segmentation of adhered seeds.

[0106] The customized U-Net network adopts an edge-weighted loss function to force the model to focus on the seed boundary pixels and solve the problem of fuzzy segmentation in the adhesion area.

[0107] Specifically, it further includes a vibration compensation module, and its motion blur correction function is:

[0108]

[0109] Where:

[0110] PSF is the point spread function obtained in real time through the IMU sensor, ε is a tiny constant to prevent division-by-zero anomalies, σ is a regularization parameter, and image blur caused by mechanical vibration is eliminated through frequency domain filtering.

[0111] The vibration compensation module collects mechanical vibration data in real time through the IMU, constructs the point spread function (PSF) and embeds it into the frequency domain filtering correction model to eliminate the motion blur caused by the movement of the 2nd flap and the 4th cylinder.

[0112] The standard parts used in the present invention can all be purchased from the market. The special-shaped parts can be customized according to the description in the specification and the drawings. The specific connection methods of each part all adopt conventional means such as bolts, rivets, and welding that are mature in the prior art. The machines, parts, and equipment all adopt conventional models in the prior art, and the circuit connection adopts the conventional connection method in the prior art, which will not be elaborated here.

[0113] In the description of the present invention, the terms "first" and "second" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, the features defined with "first" and "second" may explicitly or implicitly include one or more of such features. The meaning of "plurality" is two or more, unless otherwise specifically defined.

[0114] In the present invention, unless otherwise clearly defined and limited, the terms such as "installation", "connection", "connection", "fixation", etc. should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or integrated; it can be a mechanical connection or an electrical connection; it can be directly connected or indirectly connected through an intermediate medium, and it can be the internal communication of two components or the interaction relationship between two components. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0115] In the present invention, unless otherwise clearly defined and limited, the first feature being "on" or "under" the second feature may be that the first and second features are in direct contact, or the first and second features are indirectly in contact through an intermediate medium. Moreover, the first feature being "above", "over" and "on top of" the second feature may be that the first feature is directly above or obliquely above the second feature, or simply means that the horizontal height of the first feature is higher than that of the second feature. The first feature being "under", "below" and "beneath" the second feature may be that the first feature is directly below or obliquely below the second feature, or simply means that the horizontal height of the first feature is lower than that of the second feature.

[0116] In the description of this specification, the description with reference to terms such as "an embodiment", "some embodiments", "examples", "specific examples" or "some examples" etc. means that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in a suitable manner in any one or more embodiments or examples. In addition, without contradiction, those skilled in the art can combine and combine the different embodiments or examples described in this specification and the features of different embodiments or examples.

[0117] In the attached drawings of the disclosed embodiments of the present invention, only the structures related to the disclosed embodiments are involved, and other structures can refer to the general design. Without conflict, the same embodiment and different embodiments of the present invention can be combined with each other.

[0118] Although the present invention has been described in detail with reference to the foregoing embodiments, for those skilled in the art, they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some of the technical features. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included within the protection scope of the present invention.

Claims

1. A vegetable seed sowing device based on machine vision, characterized in that, Including: Frame (1); Two rows of flap mechanisms arranged in parallel, which are arranged inside the frame (1). Each row of flap mechanisms includes twelve independent flaps (2), and a seedling tray (3) is arranged on the top of the flap (2) for accommodating seeds. Two cylinders (4) are hinged at the bottom of the flap (2). One cylinder (4) drives the flap (2) to flip to one side for sowing qualified seeds, and the other cylinder (4) drives the flap (2) to flip to the other side for discarding unqualified seeds; Industrial vision detection component (5), which is arranged inside the frame (1) and corresponds to the two rows of flaps (2), and is used to collect seed images in each seedling tray (3) in real time and identify the number of seeds; Feeding hopper (6), which is arranged between the two rows of flaps (2) and is used to receive qualified seeds for sowing; Two return hoppers (7), which are respectively arranged outside the two rows of flaps (2) and are used to recover the unqualified seeds discarded by the flaps (2); Control system, which is arranged inside the frame (1) and is used to control the industrial vision detection component (5) and the cylinder (4).

2. The vegetable seed sowing device based on machine vision according to claim 1, characterized in that: The industrial vision detection component (5) includes a high-resolution camera and an image processing unit. The camera synchronously takes images of the seedling trays (3) of the two rows of flaps (2) from a top-down perspective, and the image processing unit counts the number of seeds through an edge detection algorithm.

3. The vegetable seed sowing device based on machine vision according to claim 2, characterized in that, The edge detection algorithm adopts an improved gradient amplitude calculation model, and the gradient operator is defined as: Where: ω x = 1 + α·var(I(x - 1:x + 1, y)), ω y = 1 + α·var(I(x, y - 1:y + 1)), α is the local texture sensitivity coefficient, var represents the gray variance of neighborhood pixels, and the edge features of high-texture regions are enhanced through dynamic weights.

4. A detection method for a vegetable seed sowing device based on machine vision, based on a vegetable seed sowing device based on machine vision according to any one of claims 1-3, characterized in that, Including the following steps: S1. Synchronously adsorb seeds into the seedling trays (3) of the two rows of flaps (2); S2. Identify the number of seeds in each seedling tray (3) through the industrial vision detection component (5) to generate detection data; S3. The control system compares the detection data with a preset threshold and performs the following operations: If at least one of the two rows of seedling trays (3) at the same longitudinal position meets the threshold, trigger the qualified flap (2) to sow and record the position of the unqualified seedling tray (3); If both do not meet the threshold, trigger the two flaps (2) to discard the seeds; S4. In the next sowing cycle, preferentially supplement and adsorb seeds into the recorded seedling trays (3) that need to be replanted.

5. The detection method of a vegetable seed sowing device based on machine vision according to claim 4, characterized in that, In S2, the multi-scale morphological reconstruction algorithm is used to identify the number of seeds, and the positioning seed region segmentation function is: Where: B k is a k-scale structural element, indicating morphological dilation, eliminating noise particles with a particle size smaller than the threshold d through iteration, min and retaining the effective seed contour.

6. The detection method of a vegetable seed sowing device based on machine vision according to claim 4, characterized in that, In S3, the preset threshold is an adaptive range threshold, and the calculation formula is: Where: A cell is the unit area of the seedling tray (3), r seed is the nominal radius of the seed, Δr is the particle size tolerance, β ∈ [0.8, 1.2] is the seeding density coefficient, and Δr is adjusted in real time through the pressure sensor at the end of the robotic arm.

7. The detection method of a vegetable seed sowing device based on machine vision according to claim 4, characterized in that, In S4, the replanting priority calculation adopts a Markov decision model, and the state transition probability is defined as: Where: s=(xmiss,tdela y ,nretry) represents the state vector of the missing position, delay time, and number of repetitions. a ∈ {immediate replanting, delayed replanting} is the action space, and the optimal replanting strategy is updated online through the Q-learning algorithm.

8. The detection method of a vegetable seed sowing device based on machine vision according to claim 5, characterized in that, Anisotropic diffusion preprocessing is performed before morphological reconstruction: Where: K = 0.3, which is the thermal conductivity, and the diffusion intensity along the long axis direction of the seed is increased by 50%, effectively maintaining the geometric integrity of the flat seed.

9. The detection method of a vegetable seed sowing device based on machine vision according to claim 8, characterized in that: After the diffusion processing, the U-Net segmentation network is integrated, and the loss function is defined as: Where: α ＝ 0.3 is the balance coefficient, BCE is the binary cross entropy, and 3 times the weight is assigned to the seed edge pixels during the calculation of the Dice coefficient, achieving an intersection over union of 92.7% for the segmentation of adherent seeds.

Images