An automatic seed analysis instrument and seed analysis method based on artificial intelligence

Abstract

The present invention relates to the field of seed testing technology, and specifically to an automatic seed testing analyzer and seed testing method based on artificial intelligence, comprising a housing, an image acquisition unit and a bright field light source arranged above the interior of the housing, wherein the image acquisition unit is signal-connected to an image processing unit and an interaction unit; wherein the image processing unit is equipped with a convolutional neural network model for extracting image features, and the convolutional neural network model is obtained by training based on historical sample image data stored in the interaction unit; the bright field light source can reduce the shadow clues of the image by adjusting its expansion degree. The identity mapping of the residual primitive is introduced, and the output of the first four levels is fused with the input of the next primitive, and then the final output is obtained through the Elu function, which effectively avoids the problem of neural network degradation and fully utilizes the features extracted by each layer of the network, solving the problem that the traditional convolutional neural network model is difficult to train.

Description

Technical Field

[0001] The present invention relates to the field of seed testing technology, and in particular to an automatic seed testing analyzer and a seed testing method based on artificial intelligence. Background Art

[0002] Seed testing involves analyzing seed cross-sections, automatically determining the number of rows, ear diameter, axis diameter, and average grain length and width along the cross-section. Traditional seed testing methods require temporary assembly of components, such as the camera, tray bracket, and light source. The connections between these components are not fixed, resulting in an unstable structure that can easily lead to measurement errors and inaccurate data. Furthermore, prior art methods only illuminate the bottom surface, creating shadowed areas during testing and causing measurement errors.

[0003] Therefore, there is an urgent need for an automatic seed analysis instrument and seed testing method that can reduce image shadows during seed testing. Summary of the Invention

[0004] The purpose of the present invention is to provide an automatic seed testing analyzer and testing method based on artificial intelligence, which is used to reduce image shadows during seed testing.

[0005] The present invention is achieved through the following technical solutions:

[0006] An automatic seed analysis instrument based on artificial intelligence includes a box, wherein a seed tray and a planar light source are provided on the bottom surface of the box. The instrument is characterized in that it also includes an image acquisition unit and a bright field light source provided above the inside of the box, wherein the image acquisition unit is signal-connected to an image processing unit and an interaction unit; wherein the image processing unit is equipped with a convolutional neural network model for extracting image features, wherein the convolutional neural network model is obtained by training based on historical sample image data stored in the interaction unit; and the bright field light source can reduce shadow clues in the image by adjusting its expansion degree. It should be noted that the seed analysis instrument in the prior art requires temporary assembly of various components, such as a camera, a bracket for fixing the tray, a light source, etc. The connection relationship between the various components is not fixed, and the structure is unstable, which can easily cause measurement errors and inaccurate measurement data. At the same time, in the prior art, only the bottom surface has a light source during seed analysis, and shadow areas exist during the seed analysis process, which can easily cause measurement errors.

[0007] To address these challenges, an artificial intelligence-based automatic seed analyzer was proposed. By installing a planar light source and a bright-field light source within the instrument's housing, and activating both simultaneously during seed analysis, the instrument eliminates shadows and reduces image acquisition errors. Furthermore, the analyzer needs to be flexible and adaptable to different seed types, such as rice, wheat, rapeseed, and soybeans, while also accounting for insect population analysis, insect counts on smooth surfaces, and egg counts. Therefore, dynamic adjustment of the light source is necessary for different usage conditions. The bright-field light source can be adjusted to varying degrees of expansion and contraction, thereby macro-controlling image quality. The image processing unit incorporates a convolutional neural network model, which is used to extract network features from images at different scales. Conventional convolutional neural network models improve the level of feature information they can extract as the number of network layers increases. However, this also leads to network degradation. Therefore, balancing feature extraction and minimizing degradation is crucial. Based on this problem, the applicant proposed a convolutional neural network model for super-resolution, which greatly avoids the degradation problem of neural networks and makes full use of the features extracted by each layer of the network, thereby enabling the convolutional neural network model to better exhibit local feature extraction characteristics.

[0008] Furthermore, the convolutional neural network model includes multiple extraction primitives, and the extraction primitives include four convolution layers, the first layer is a 1*1 convolution kernel, the second layer is a 3*3 convolution kernel, the third layer is a 1*1 convolution kernel, and the fourth layer is a 1*1 convolution kernel, wherein the stride of the first three layers is 1, the stride of the fourth layer is set by the input of the interaction unit, and the four convolution layers are connected through a dense network. It should be noted that for this extraction unit, the identity mapping of the residual primitive is introduced, and the output of the first four levels is fused with the input of the next primitive, and then the final output is obtained by the Elu function. Based on the above content, the problem of neural network degradation is effectively avoided, and the features extracted by each layer of the network are fully utilized, solving the problem that the traditional convolutional neural network model is difficult to train.

[0009] Furthermore, the four convolutional layers use Elu as the activation function, and the output of each convolutional layer is fused with the input of the next convolutional layer, and finally the final output is obtained after passing through the Elu function. By connecting several extraction primitives at the end and fusing them together, the convolutional neural network model can extract higher-level information from the image, making better use of the information extracted by each module and extracting more detailed local image features.

[0010] Furthermore, the bright field light source includes: a fixed base, a plurality of expansion sections, and an expansion member. The fixed base is fixedly connected to the upper end surface of the interior of the housing. The plurality of expansion sections are hingedly arranged in a circular array on the outer circumference of the fixed base. The expansion member is hingedly connected to the expansion section, and the expansion section can be opened and closed by the expansion member. The lower end of the expansion section is provided with a surface light source. It should be noted that based on the above structure, the expansion member adjusts the degree of expansion of the expansion section on the outer circumference of the fixed base to adjust the illumination area and brightness of the light source. The expansion operation is carried out synchronously, which can effectively avoid the impact of traditional fixed light sources on seeds of different sizes and indirectly improve the accuracy of seed testing.

[0011] Furthermore, the unfolding part includes: a limit frame fixedly arranged on the fixing seat, and an unfolding bracket hingedly arranged on the limit frame. The bright field light source also includes: a motor arranged in the fixing seat, two symmetrically arranged limit plates, and an unfolding plate arranged between the two limit plates. The output end of the motor is connected to the unfolding plate, and a number of unfolding grooves are evenly distributed on the unfolding plate. A number of limit grooves are evenly distributed on the limit plate. The end of the unfolding bracket is also hinged with a transmission rod, and the free end of the transmission rod passes through the limit slot and cooperates with the unfolding slot. When the motor is started, the opening and closing of the unfolding bracket is realized by driving the limit plate to rotate so that the transmission rod moves in the limit slot. It should be noted that the bright limit plate is fixed to the limit frame and is located at the center of the limit frame. The expansion plate is rotatably arranged between the two limit plates. When the motor is started, it can drive the expansion plate to rotate between the limit plates. It should also be noted that the rotation of the limit plate can drive the reciprocating motion of the end of the transmission rod in the limit groove, that is, the reciprocating motion of the other end of the transmission rod in the limit hole of the limit frame is realized, thereby realizing the opening and closing process of the expansion bracket. Since the expansion bracket is also hinged to the expansion part, when the expansion bracket is opened and closed, the expansion part opens and closes accordingly, and the surface light source arranged in the lower end of the expansion part moves together.

[0012] Furthermore, a limiting hole is provided on the limiting frame, one end of the limiting frame and one end of the transmission rod are hinged in the limiting hole, the other end of the limiting frame is hinged to the bottom of the limiting frame, and the limiting frame is also hinged to the unfolding portion, and the other end of the transmission rod is provided with a limiting slide, and the limiting slide matches the unfolding slot and the limiting slot at the same time. It should be noted that the limiting slide can move in the unfolding slot and the limiting slot. Since the unfolding slot and the limiting slot have different shapes, when the unfolding plate rotates between the two limiting plates, it can drive the limiting slide to move in the limiting slot. It should also be noted that since the unfolding slots are all opened on the same unfolding plate, the unfolding steps of all the unfolding parts are in unison and can move together.

[0013] Preferably, a reflective layer is provided between the unfolding parts to avoid mutual influence between the unfolding parts during closure.

[0014] A seed testing method includes the following steps: step 1, collecting image data through an image acquisition unit and transmitting the image data to an image processing unit; step 2, after the image processing unit receives the image data in step 1, performing grayscale transformation on the image; step 3, performing binary threshold segmentation on the image after the grayscale transformation in step 2; step 4, a convolutional neural network model in the image processing unit calls historical samples stored in an interaction unit, establishes a deep learning sample set, and uses the deep learning sample set for training; step 5, inputs the image segmented in step 3 into the trained convolutional neural network model, and automatically processes the image data through the convolutional neural network model, thereby performing a seed testing operation.

[0015] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0016] 1. The bright field light source and convolutional neural network model of the present invention can jointly process image features, reducing image shadows during seed testing;

[0017] 2. The present invention introduces the identity mapping of the residual primitive, and fuses the output of the first four levels with the input of the next primitive, and then obtains the final output through the Elu function. Based on this content, the problem of neural network degradation is effectively avoided, and the features extracted by each layer of the network are fully utilized, solving the problem that the traditional convolutional neural network model is difficult to train;

[0018] 3. The present invention adjusts the degree of expansion of the expansion part on the outer peripheral surface of the fixed seat through the expansion piece to adjust the irradiation area and brightness of the light source. The expansion operation is carried out synchronously, which can effectively avoid the influence of traditional fixed light sources on seeds of different sizes and indirectly improve the seed testing accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The drawings described herein are used to provide a further understanding of the embodiments of the present invention, constitute a part of this application, and do not constitute a limitation of the embodiments of the present invention. In the drawings:

[0020] Figure 1 It is a structural schematic diagram of the present invention;

[0021] Figure 2 Schematic diagram of the structure of the bright field light source;

[0022] Figure 3 Schematic diagram of the structure of the bright field light source in the expanded state;

[0023] Figure 4This is a schematic diagram of the structure of a bright field light source in an upward-looking state;

[0024] Figure 5 Schematic diagram of the internal structure of the bright field light source;

[0025] Figure 6 for Figure 5 Schematic diagram of the enlarged structure of A;

[0026] Figure 7 for Figure 5 Schematic diagram of the enlarged structure of B;

[0027] Figure 8 Schematic diagram of the structure of the unfolded plate;

[0028] Figure 9 A schematic diagram of the structure for extracting primitives;

[0029] Figure 10 It is a flowchart of the method of the present invention.

[0030] Markings and corresponding parts names in the accompanying drawings:

[0031] 1-box, 2-test tray, 3-bright field light source, 31-fixed seat, 32-expansion part, 33-expansion part, 34-reflective layer, 35-limiting plate, 36-expansion plate, 37-expansion slot, 38-limiting slot, 331-limiting frame, 332-expansion bracket, 333-transmission rod, 334-limiting hole, 335-limiting slide. DETAILED DESCRIPTION

[0032] To make the objectives, technical solutions, and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the examples and accompanying drawings. The exemplary embodiments of the present invention and their descriptions are only used to explain the present invention and are not intended to limit the present invention. It should be noted that the present invention is already in the actual development and use stage.

[0033] Example 1:

[0034] Please refer to the attached Figures 1 to 9As shown, an automatic seed analysis instrument based on artificial intelligence includes a box body 1, wherein a seed tray 2 and a planar light source are provided on the bottom surface of the box body 1. The instrument is characterized in that it also includes an image acquisition unit and a bright field light source 3 disposed above the interior of the box body 1, wherein the image acquisition unit is signal-connected to an image processing unit and an interaction unit; wherein the image processing unit is equipped with a convolutional neural network model for extracting image features, and the convolutional neural network model is obtained by training based on historical sample image data stored in the interaction unit; and the bright field light source 3 can reduce shadow clues in the image by adjusting its expansion degree. It should be noted that seed analysis instruments in the prior art require temporary assembly of various components, such as a camera, a bracket for fixing the tray, a light source, etc. The connection relationship between the various components is not fixed, and the structure is unstable, which is prone to measurement errors and inaccurate measurement data. At the same time, in the prior art, only the bottom surface has a light source during seed analysis, and shadow areas exist during the seed analysis process, which is prone to measurement errors.

[0035] To address the above-mentioned issues, an artificial intelligence-based automatic seed analysis instrument is proposed. By disposing a planar light source and a bright field light source 3 within a housing 1, and activating both simultaneously during seed analysis, shadows can be eliminated and image acquisition errors reduced. Furthermore, the instrument needs to be flexible and adaptable to different seed types, such as rice, wheat, rapeseed, and soybeans, and must also consider insect population analysis, insect counts on smooth surfaces, and egg counts. Therefore, dynamic adjustment of the light source is required for different usage conditions. The bright field light source 3 can be adjusted by varying its expansion and contraction to adjust its brightness, thereby macro-controlling image quality. The image processing unit incorporates a convolutional neural network model, which is used to extract network features from images at different scales. Conventional convolutional neural network models improve the level of feature information they can extract as the number of network layers increases. However, as the number of network layers increases, the network degrades. Therefore, balancing feature information extraction and mitigating degradation is particularly important. Based on this problem, the applicant proposed a convolutional neural network model for super-resolution, which greatly avoids the degradation problem of neural networks and makes full use of the features extracted by each layer of the network, thereby enabling the convolutional neural network model to better exhibit local feature extraction characteristics.

[0036] It should be noted that the convolutional neural network model includes multiple extraction primitives, and the extraction primitives include four convolution layers, the first layer is a 1*1 convolution kernel, the second layer is a 3*3 convolution kernel, the third layer is a 1*1 convolution kernel, and the fourth layer is a 1*1 convolution kernel, wherein the stride of the first three layers is 1, the stride of the fourth layer is set by the input of the interaction unit, and the four convolution layers are connected through a dense network. It should also be noted that for this extraction unit, the identity mapping of the residual primitive is introduced, and the output of the first four levels is fused with the input of the next primitive, and then the final output is obtained by the Elu function. Based on the above content, the problem of neural network degradation is effectively avoided, and the features extracted by each layer of the network are fully utilized, solving the problem that the traditional convolutional neural network model is difficult to train.

[0037] It should be noted that the four convolutional layers use Elu as the activation function, and the output of each convolutional layer is fused with the input of the next convolutional layer, and finally the final output is obtained after passing through the Elu function. By connecting several extraction primitives and fusing them together, the convolutional neural network model can extract higher-level information from the image, making better use of the information extracted by each module and extracting more detailed local image features.

[0038] The encoder part of the convolutional neural network is divided into a global feature extraction module and a local feature extraction module. The global feature extraction model consists of a 7*7 convolution layer and a pooling layer. The local feature extraction module consists of four end-to-end extraction primitives. The encoder first uses the global feature extraction module to globally extract the entire image information, and then uses the local feature extraction module to perform secondary extraction of the image's detailed features.

[0039] It should be noted that the bright field light source 3 includes: a fixed seat 31, a plurality of expansion parts 32 and an expansion member 33. The fixed seat 31 is fixedly connected to the upper end face inside the box body 1. The plurality of expansion parts 32 are hingedly arranged on the outer peripheral surface of the fixed seat 31 in a circular array. The expansion member 33 is hinged to the expansion part 32. The expansion part 32 can be opened and closed by the expansion member 33, and a surface light source is provided at the lower end of the expansion part 32. It should also be noted that based on the above structure, the irradiation area and brightness of the light source are adjusted by adjusting the expansion degree of the expansion part 32 on the outer peripheral surface of the fixed seat 31 through the expansion member 33. The expansion operation is carried out synchronously, which can effectively avoid the influence of the traditional fixed light source on seeds of different sizes and indirectly improve the seed testing accuracy. The closed state of the expansion part 32 is shown in the attached figure. Figure 2 As shown, the overall structure is a circular cover; the opening and closing state is as shown in the attached Figure 3 As shown, the overall structure is an open flower.

[0040] It should be noted that the unfolding member 33 includes: a limit frame 331 fixedly arranged on the fixing seat 31, and an unfolding bracket 332 hingedly arranged on the limit frame 331. The bright field light source 3 also includes: a motor arranged in the fixing seat 31, two symmetrically arranged limit plates 35, and an unfolding plate 36 arranged between the two limit plates 35. The output end of the motor is connected to the unfolding plate 36, and a number of unfolding grooves 37 are evenly spaced on the unfolding plate 36. A number of limit grooves 38 are evenly spaced on the limit plate 35. The end of the unfolding bracket 332 is also hinged with a transmission rod 333, and the free end of the transmission rod 333 passes through the limit groove 38 and cooperates with the unfolding groove 37. When the motor is started, the opening and closing of the unfolding bracket 332 is realized by driving the limit plate 35 to rotate so that the transmission rod 333 moves in the limit groove 38. It should be noted that the bright limit plate 35 is fixed to the limit frame 331 and is located at the center of the limit frame 331. The expansion plate 36 is rotatably arranged between the two limit plates 35. When the motor is started, it can drive the expansion plate 36 to rotate between the limit plates 35. It should also be noted that the rotation of the limit plate 35 can drive the end of the transmission rod 333 to reciprocate in the limit slot 38, that is, to achieve the reciprocating movement of the other end of the transmission rod 333 in the upper limit hole 334 of the limit frame 331, thereby achieving the opening and closing process of the expansion bracket 332. Since the expansion bracket 332 is also hinged to the expansion part 32, when the expansion bracket 332 opens and closes, the expansion part 32 also opens and closes, and the surface light source set in the lower end of the expansion part 32 moves together. For the limit slot 38, its structure is preferably a straight slot, for the expansion slot 37, its structure is preferably an arc hole, and for the limit hole 334, it is also preferably an arc hole. The number of the deployment members 33 is preferably 8, and the number of the limiting slots 38, the deployment slots 37, the limiting holes 334, etc. is corresponding. As for the deployment bracket 332, it is preferably a butterfly bracket mechanism.

[0041] It should be noted that the limiting bracket 331 has a limiting hole 334. One end of the limiting bracket is hingedly connected to one end of the transmission rod 333 within the limiting hole 334. The other end of the limiting bracket is hingedly connected to the bottom of the limiting bracket 331 and is also hingedly connected to the deployment portion 32. The other end of the transmission rod 333 is provided with a limiting slide 335, which mates with the deployment slot 37 and the limiting slot 38. It should be noted that the limiting slide 335 can move within the deployment slot 37 and the limiting slot 38. Due to the different shapes of the deployment slot 37 and the limiting slot 38, when the deployment plate 36 rotates between the two limiting plates 35, it can drive the limiting slide 335 to move within the limiting slot 38. It should also be noted that because the deployment slots 37 are all provided on the same deployment plate 36, all deployment members 33 deploy in unison and can move together.

[0042] Preferably, a reflective layer 34 is provided between the unfolding parts 32. The reflective layer 34 is provided to avoid mutual influence between the unfolding parts 32 when closing.

[0043] Example 2:

[0044] This embodiment only describes the parts that differ from the first embodiment, specifically:

[0045] As attached Figure 10 As shown, a testing method includes the following steps: step 1, collecting image data through an image acquisition unit, and transferring the image data to an image processing unit; step 2, after the image processing unit receives the image data in step 1, grayscale conversion is performed on the image; step 3, the image that has completed the grayscale conversion in step 2 is subjected to binarization threshold segmentation; step 4, the convolutional neural network model in the image processing unit calls the historical samples stored in the interaction unit, and establishes a deep learning sample set, and uses the deep learning sample set for training; step 5, inputting the image after segmentation in step 3 into the trained convolutional neural network model, and automatically processing the image data through the convolutional neural network model, thereby performing the testing operation. Based on the above method, the convolutional neural network model can be used to learn features from the sample set of historical data, thereby quickly extracting image features, and has strong adaptability and robustness.

[0046] Example 3:

[0047] This embodiment only describes the parts that differ from the first embodiment, specifically:

[0048] The problem of image shadow acquisition requires targeted definition of the target's edge pixels. Specifically, the detected target edge pixels are first classified to determine whether the pixel is a line segment endpoint. The specific judgment method is to search the eight neighborhood pixels around the pixel (establish a neighborhood topology map). If there is a pixel connected to the pixel, then the point is a line segment endpoint and is saved in the endpoint array; otherwise, it is not a line segment endpoint. For non-line segment endpoint pixels, they are divided into isolated points, a point in the edge line (two pixels in the eight neighborhoods are connected to it), and an edge line intersection point (four pixels in the eight neighborhoods are connected to it) according to their connectivity.

[0049] Then, the BFS algorithm is used to search the pixel points in the endpoint array until the search stop condition is met. The search path is marked as closed, and then the scan is restarted from the interruption point. The best subsequent endpoint (which does not belong to the same connected component as the current pixel point) is selected for connection, and finally the complete hilum and kidney edge that meets the requirements is obtained.

[0050] The specific steps are as follows:

[0051] (1) Take the edge pixel of the image currently being searched as the search starting point and perform the first-level search. If the endpoint of the line segment is found, the search ends and the above search is repeated for the next pixel; otherwise, go to step (2);

[0052] (2) If the target point cannot be found in step 1, it is necessary to expand the search for each pixel point found in the first layer one by one, repeat the search for each point until the endpoint of the line segment is found, and then repeat the same steps for the next pixel point; otherwise, it is necessary to determine whether the number of search layers exceeds the preset threshold. If it exceeds the threshold, the search ends and the BFS search continues for the next image edge pixel point. If it does not exceed the threshold, the second layer of pixels is further expanded and the process returns to step (1) and is repeated.

[0053] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. An automatic seed analysis instrument based on artificial intelligence, comprising a box (1), wherein a seed analysis tray (2) and a plane light source are provided on the bottom surface of the box (1), characterized in that: It also includes an image acquisition unit and a bright field light source (3) arranged above the interior of the box (1), wherein the image acquisition unit is signal-connected to the image processing unit and the interaction unit; The image processing unit is equipped with a convolutional neural network model for extracting image features, and the convolutional neural network model is obtained after training based on the historical sample image data stored in the interaction unit; The bright field light source (3) can reduce shadow clues in the image by adjusting its expansion degree; The convolutional neural network model includes a plurality of extraction primitives, and the extraction primitives include four convolution layers, the first layer is a 1*1 convolution kernel, the second layer is a 3*3 convolution kernel, the third layer is a 1*1 convolution kernel, and the fourth layer is a 1*1 convolution kernel, wherein the stride of the first three layers is 1, the stride of the fourth layer is set by the input of the interaction unit, and the four convolution layers are connected through a dense network; the bright field light source (3) includes: a fixed seat (31), a plurality of expansion parts (32) and an expansion member (33), the fixed seat (31) is fixedly connected to the upper end surface inside the box (1), the plurality of expansion parts (32) are hingedly arranged on the outer peripheral surface of the fixed seat (31) in a circular array, the expansion member (33) is hinged to the expansion part (32), the expansion part (32) can be opened and closed by the expansion member (33), and a surface light source is arranged at the lower end of the expansion part (32).

2. The automatic seed analysis instrument based on artificial intelligence according to claim 1, characterized in that: The four convolutional layers use Elu as the activation function, and the output of each convolutional layer is fused with the input of the next convolutional layer, and finally the final output is obtained after passing through the Elu function.

3. The automatic seed analysis instrument based on artificial intelligence according to claim 1, characterized in that: The unfolding member (33) comprises: a limiting frame (331) fixedly arranged on the fixing seat (31), and an unfolding bracket (332) hingedly arranged on the limiting frame (331). The bright field light source (3) further comprises: a motor arranged in the fixing seat (31), two symmetrically arranged limiting plates (35), and an unfolding plate (36) arranged between the two limiting plates (35). The output end of the motor is connected to the unfolding plate (36). The unfolding plate (36) is evenly spaced with a plurality of unfolding grooves (37). The limiting plate (35) is evenly spaced with a plurality of limiting grooves (38). The end of the unfolding bracket (332) is also hingedly connected to a transmission rod (333). The free end of the transmission rod (333) passes through the limiting groove (38) and cooperates with the unfolding groove (37). When the motor is started, the transmission rod (333) moves in the limiting groove (38) by driving the limiting plate (35) to rotate, thereby realizing the opening and closing of the unfolding bracket (332).

4. The automatic seed analysis instrument based on artificial intelligence according to claim 3, characterized in that: A limiting hole (334) is provided on the limiting frame (331), one end of the limiting frame and one end of the transmission rod (333) are hinged in the limiting hole (334), the other end of the limiting frame is hinged to the bottom of the limiting frame (331), and the limiting frame is also hinged to the unfolding portion (32), and a limiting slide (335) is provided at the other end of the transmission rod (333), and the limiting slide (335) is matched with the unfolding slot (37) and the limiting slot (38) at the same time.

5. The automatic seed analysis instrument based on artificial intelligence according to claim 1, characterized in that: A reflective layer (34) is also provided between the unfolding portions (32).

6. A method for testing a seed, characterized in that: An artificial intelligence-based automatic seed analysis instrument according to any one of claims 1 to 5, comprising the following steps: Step 1: collecting image data through the image acquisition unit and transmitting the image data to the image processing unit; Step 2: After receiving the image data in step 1, the image processing unit performs grayscale conversion on the image; Step 3, performing binary threshold segmentation on the image that has completed the grayscale transformation in step 2; Step 4: The convolutional neural network model in the image processing unit calls the historical samples stored in the interaction unit, establishes a deep learning sample set, and uses the deep learning sample set for training; In step 5, the image segmented in step 3 is input into the trained convolutional neural network model, and the convolutional neural network model is used to automatically process the image data, thereby performing a test operation.

Images