High-throughput DUS test method and device and storage medium

Abstract

The invention provides a high-throughput DUS test method and device and a storage medium. The high-throughput DUS test method comprises the steps of performing semantic segmentation on a to-be-processed image of a target based on a target segmentation model to obtain a semantic segmentation result; the target segmentation model is a semantic segmentation model constructed based on a dense connection void space pyramid pooling DenseASPP module and a multi-scale dilatation attention MSDA module; based on the semantic segmentation result, utilizing a high-throughput phenotypic technology to obtain phenotypic characters of the target; and determining the variety of the to-be-identified crop based on the phenotypic character. According to the high-throughput DUS testing method and device and the storage medium, the automation degree of DUS testing is remarkably improved, and the DUS testing efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and in particular, to a high-throughput DUS testing method, apparatus, and storage medium. Background Art

[0002] With the rapid development of agricultural science and technology, the breeding and protection of new plant varieties have become one of the important directions in agricultural research. To ensure the intellectual property rights and market competitiveness of new varieties, it is necessary to conduct DUS (Distinctness, Uniformity, and Stability) testing on new plant varieties for detection and certification. DUS testing plays an important role in the protection of new plant varieties and is particularly indispensable in the identification and registration of germplasm resources such as horticultural crops like lettuce.

[0003] Traditional DUS testing mainly relies on manual observation and measurement, including qualitative description and quantitative evaluation of plant appearance characteristics. These characteristics usually include the shape, color, edge contour, texture, etc. of the leaves. However, the manual observation method consumes too much time and manpower, and the testing efficiency is low. Summary of the Invention

[0004] The present invention provides a high-throughput DUS testing method, apparatus, and storage medium to solve the technical problem of low DUS testing efficiency in the prior art.

[0005] In a first aspect, the present invention provides a high-throughput DUS testing method, including the following steps.

[0006] Performing semantic segmentation on a to-be-processed image of a target based on a target segmentation model to obtain a semantic segmentation result; the target segmentation model is a semantic segmentation model constructed based on a DenseASPP (Dense Connected Atrous Spatial Pyramid Pooling) module and a multi-scale dilated attention (MSDA) module; Obtaining phenotypic traits of the target by using high-throughput phenotyping technology based on the semantic segmentation result; Determining the variety of the crop to be identified based on the phenotypic traits.

[0007] In some embodiments, the determining the variety of the crop to be identified based on the phenotypic traits includes: Obtaining variety fingerprints of different varieties based on the phenotypic traits of different varieties; Determining the variety of the crop to be identified based on the variety fingerprints and the image of the crop to be identified.

[0008] In some embodiments, the determining the variety of the crop to be identified based on the variety fingerprints and the image of the crop to be identified includes: Obtaining the phenotypic traits of the crop to be identified based on the image of the crop to be identified; The variety of the crop to be identified is determined by comparing the phenotypic traits of the crop to be identified with the variety fingerprint.

[0009] In some embodiments, obtaining the phenotypic traits of the target by using the high-throughput phenotyping technology based on the semantic segmentation result includes: A normalized semantic segmentation result is obtained by rotating and scaling the semantic segmentation result. The basic phenotypic parameters of the target are extracted from the normalized semantic segmentation result by using the high-throughput phenotyping technology. The phenotypic traits of the target are quantified according to the basic phenotypic parameters.

[0010] In some embodiments, the target segmentation model is a semantic segmentation model DeepLabV3+ improved based on the DenseASPP module and the MSDA module, and the backbone network of the improved DeepLabV3+ is an efficient ShuffleNetV2 network constructed by lightweight depthwise separable convolution and channel shuffle operations.

[0011] In some embodiments, the method further includes: Collecting target images of different varieties of crops. Performing data annotation and data augmentation on the target images to obtain the images to be processed of the target.

[0012] In a second aspect, the present invention provides a high-throughput DUS testing device, including the following modules.

[0013] A first acquisition module, configured to perform semantic segmentation on the image to be processed of the target based on a target segmentation model to obtain a semantic segmentation result; the target segmentation model is a semantic segmentation model constructed based on a dense connection atrous spatial pyramid pooling DenseASPP module and a multi-scale dilated attention MSDA module. A second acquisition module, configured to obtain the phenotypic traits of the target by using the high-throughput phenotyping technology based on the semantic segmentation result. A determination module, configured to determine the variety of the crop to be identified based on the phenotypic traits.

[0014] In a third aspect, the present invention provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, the high-throughput DUS testing method as described in any one of the above is implemented.

[0015] In a fourth aspect, a non-transitory computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the high-throughput DUS testing method as described in any one of the above is implemented.

[0016] In a fifth aspect, the present invention also provides a computer program product, including a computer program which, when executed by a processor, implements the high-throughput DUS testing method as described in the first aspect above.

[0017] The high-throughput DUS testing method, device and storage medium provided by the present invention perform semantic segmentation on a to-be-processed image of an object based on an object segmentation model to obtain a semantic segmentation result. The object segmentation model is a semantic segmentation model constructed based on a DenseASPP (Dense Connected Atrous Spatial Pyramid Pooling) module and a MSDA (Multi-Scale Dilated Attention) module. Based on the semantic segmentation result, the phenotypic traits of the object are obtained by using high-throughput phenotyping technology, and the variety of the crop to be identified is determined based on the phenotypic traits, significantly improving the automation degree of DUS testing and the DUS testing efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0019] Figure 1 It is a flowchart of the high-throughput DUS testing method provided by the present invention.

[0020] Figure 2 It is a structural diagram of the high-throughput DUS testing device provided by the present invention.

[0021] Figure 3 It is a structural diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0022] The traditional manual observation method has the following defects: (1) Time-consuming and laborious: Manual operation usually requires experts to observe and record one by one, which is not only a cumbersome process but also requires a large amount of time and human resources.

[0023] (2) Strong subjectivity: Since human vision and judgment are affected by factors such as environment, experience, and emotion, it often leads to subjective deviation and insufficient consistency of measurement results.

[0024] (3) Low efficiency: The traditional testing method is difficult to process a large number of samples in a short time, especially in large-scale planting and high-density breeding projects, and the efficiency problem is particularly prominent.

[0025] To address the problems in traditional DUS test traits, high-throughput phenotyping (HTP) technology has gradually been applied to the fields of plant breeding and phenotypic trait analysis. The introduction of this technology makes it possible to accurately measure and rapidly analyze plant phenotypic traits.

[0026] (1) Non-destructive measurement: Non-contact measurement is carried out using cameras and sensors, reducing damage to plants and ensuring the integrity of samples.

[0027] (2) Multi-source data acquisition: High-throughput technology can simultaneously obtain multi-dimensional information of plants, including RGB images, spectral information (such as near-infrared light NIR), 3D structures, etc., providing more comprehensive data support for phenotypic analysis.

[0028] (3) Automation and high efficiency: Through automated equipment and data processing algorithms, a large number of samples can be processed in a short time, improving work efficiency.

[0029] Although high-throughput phenotyping technology shows great potential in plant phenotypic analysis, there are still technical bottlenecks in its specific application in DUS testing: (1) Complex data analysis: The data generated by high-throughput phenotyping technology is huge. How to extract discriminative DUS test trait features from it remains a challenge.

[0030] (2) Insufficient universality and robustness of models: Some existing machine learning and deep learning models often rely on a large amount of labeled data, and the acquisition and labeling of data require a large amount of human and time resources. In addition, the universality and robustness of these models among different environments and varieties still need to be improved.

[0031] (3) Lack of standardized cultivar fingerprint construction: There is currently no standardized process and algorithm to convert the phenotypic characteristics of lettuce leaves into fingerprint information for cultivar identification.

[0032] Based on the above technical problems, the present invention proposes a high-throughput DUS test method. A semantic segmentation model constructed based on the DenseASPP module and the MSDA module is used to perform semantic segmentation on the image to be processed, improving the accuracy of semantic segmentation. The phenotypic traits of the target are obtained through high-throughput phenotyping technology, ensuring the effectiveness of determining DUS test traits, significantly improving the automation degree, precision, and efficiency of DUS testing, and being applicable to large-scale cultivar identification and new cultivar testing.

[0033] To make the objectives, technical solutions, and advantages of the present invention clearer, the technical solutions in the present invention will be clearly and completely described below with reference to the accompanying drawings in the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present invention without creative efforts shall fall within the protection scope of the present invention.

[0034] Figure 1 is one of the schematic flowcharts of the high-throughput DUS testing method provided by the present invention. As Figure 1 shown, the present invention provides a high-throughput DUS testing method. The method includes: Step 101: Perform semantic segmentation on the image to be processed of the target based on the target segmentation model to obtain a semantic segmentation result; the target segmentation model is a semantic segmentation model constructed based on the Dense Atrous Spatial Pyramid Pooling (DenseASPP) module and the Multi-Scale Dilated Attention (MSDA) module.

[0035] Specifically, the target can be a part that can distinguish crop varieties, such as a lettuce leaf. The image to be processed of the target can be a preprocessed target image, such as a preprocessed lettuce leaf image.

[0036] Perform semantic segmentation on the image to be processed of the target using the target segmentation model, where the target segmentation model is a semantic segmentation model constructed based on the Dense Atrous Spatial Pyramid Pooling (DenseASPP) module and the Multi-Scale Dilated Attention (MSDA) module.

[0037] Among them, the DenseASPP module can extract denser pixel points and a larger receptive field range. The MSDA module can not only capture local details but also perceive context information in a wider area, improving the accuracy of semantic segmentation.

[0038] Step 102: Obtain the phenotypic traits of the target using high-throughput phenotyping technology based on the semantic segmentation result.

[0039] Specifically, high-throughput phenotyping technology (HTP) refers to the rapid and non-destructive acquisition of various phenotypic traits (phenotypic characters) of plants, such as morphology, color, texture, and spectral information, through a variety of sensors and automated devices.

[0040] After obtaining the semantic segmentation result, using the high-throughput phenotyping technology, that is, constructing a high-throughput phenotypic trait extraction pipeline to extract the phenotypic traits of targets such as lettuce leaves. Among them, the phenotypic traits of lettuce leaves can include shape traits such as leaf shape, leaf tip shape, leaf margin shape, and vein shape, and color traits such as leaf hue, color lightness, and anthocyanin content.

[0041] Step 103: Determine the variety of the crop to be identified based on the phenotypic traits.

[0042] Specifically, the obtained phenotypic traits are the phenotypic traits corresponding to different varieties of a certain crop. In practical applications, the variety of the crop to be identified can be identified based on the phenotypic traits corresponding to each variety obtained in advance.

[0043] For example, obtain the phenotypic traits corresponding to each variety of lettuce, including phenotypic trait 1 corresponding to lettuce variety 1, phenotypic trait 2 corresponding to lettuce variety 2, phenotypic trait 3 corresponding to lettuce variety 3, and so on. Now, an image of the lettuce to be identified is collected, and the phenotypic traits of the lettuce to be identified are analyzed. If the phenotypic traits of the lettuce to be identified are consistent with phenotypic trait 2, it is determined that the lettuce to be identified belongs to lettuce variety 2.

[0044] The high-throughput DUS testing method provided by the embodiments of the present application uses a semantic segmentation model constructed based on the DenseASPP module and the MSDA module to perform semantic segmentation on the image to be processed, improving the accuracy of semantic segmentation. By using the high-throughput phenotyping technology, the effectiveness of DUS test trait determination is ensured, realizing fast and accurate DUS testing, which is applicable to large-scale variety identification and new variety testing.

[0045] In some embodiments, the method further includes: Collect the target image; Perform data annotation and data augmentation on the target image to obtain the image to be processed of the target.

[0046] Specifically, take the target image as an example of lettuce leaf images. In a certain greenhouse, 303 lettuce varieties are planted in pots, including Butter, Crisphead, Leaf, Oakleaf, Roman, Stem, and Wild relatives. 30 days after planting, the seedlings are transplanted into flowerpots with a diameter of 32 cm and a height of 34 cm. There are 3 plants for each variety, arranged in a row, and normal water and fertilizer management is carried out. At the mature stage, 270 lettuce varieties (810 plants) are selected. Near the greenhouse, a photography studio is set up, and the images are taken with a background cloth, a tripod, and a digital camera. A 22 mm fixed-focus lens is used to obtain images at a distance of about 1.2 m. During the data collection process, the most representative leaf is cut from each plant and immediately used to obtain two images of the front and back sides. At the same time, a color colorimetric card is placed near the leaf for later calibration. The image resolution is set to 40 million pixels (5328×4000), the camera is set to the program automatic exposure mode, and the images are stored in JPG format, with a total of 1620 lettuce leaf images.

[0047] After obtaining the target image and before performing semantic segmentation based on the target segmentation model, it is necessary to perform data annotation and data augmentation on the target image to obtain the image to be processed.

[0048] For example, use the image annotation software Labelme to manually annotate the data, including leaf blade, main vein, lateral vein, etc., annotate 101 representative lettuce leaves, and use data augmentation techniques such as rotation, flipping, scaling, cropping, translation, and affine transformation for data augmentation, expanding to 808, to obtain the image to be processed.

[0049] The high-throughput DUS test method provided by the embodiments of the present application enriches the data for model training through data augmentation, reduces the risk of overfitting, and improves the generalization ability of the model.

[0050] In some embodiments, the target segmentation model is the semantic segmentation model DeepLabV3+ improved based on the DenseASPP module and the MSDA module, and the backbone network of the improved DeepLabV3+ is the lightweight convolutional neural network ShuffleNetV2.

[0051] Specifically, the target segmentation model in the embodiments of the present application can be a semantic segmentation model based on DeepLabV3+. For the semantic segmentation model DeepLabV3+ with an encoder-decoder structure, which performs end-to-end semantic segmentation by adopting a DCNN backbone network and a pyramid pooling module, there are problems of high computational complexity and large memory consumption. Moreover, when extracting image feature information, it cannot make full use of multi-scale information, easily causing the loss of detailed information and resulting in impaired segmentation accuracy. To further improve the segmentation performance, the DeepLabV3+ network can be improved. Specifically, it is improved based on the DenseASPP module and the MSDA module, and the backbone network is replaced with an efficient ShuffleNetV2 network constructed by lightweight depthwise separable convolution and channel shuffle operations.

[0052] First, since the number of network parameters in the feature extraction network of the original deep convolutional neural network (DCNN) backbone network is too large, the embodiments of the present application use an efficient ShuffleNetV2 network constructed by lightweight depthwise separable convolution and channel shuffle operations. The channel shuffle operation is introduced to achieve efficient information transmission by "shuffling" the features of different channels, ensuring information flow and representation ability. The depthwise separable convolution is introduced to separate the spatial convolution and the channel convolution, reducing the amount of computation and the number of parameters and improving the efficiency. Residual connections are introduced to avoid gradient vanishing and accelerate training, allowing information to be directly transmitted through skip connections, accelerating the training process of the network and improving the network performance.

[0053] Then, DeepLabV3+ extracts multi-scale feature receptive fields by stacking atrous pyramid pooling modules with different dilation rates in parallel. In theory, to obtain a sufficiently large receptive field, a larger dilation rate needs to be increased. However, as the dilation rate increases, the decay of the atrous convolution becomes ineffective. The embodiments of the present application instead use a densely connected atrous spatial pyramid pooling DenseASPP module to extract denser pixel points and a larger receptive field range.

[0054] Finally, to enhance the ability of the feature map to extract detailed information, improve the accuracy of semantic segmentation, and then restore clearer segmentation boundaries, the MSDA module is introduced after the DenseASPP module and after the underlying features of ShuffleNetV2 respectively. By using the self-attention mechanism of different heads, semantic information at different levels and multi-scales is captured. By separating and parallel processing the channels of the feature map, the learning ability and efficiency are enhanced, and different dilation rate heads are used to focus on features at different scales to comprehensively capture key information. Finally, a richer feature representation is obtained through linear aggregation operations to improve the segmentation performance.

[0055] The high-throughput DUS testing method provided by the embodiments of the present application improves DeepLabV3+ based on ShuffleNetV2, DenseASPP module, and MSDA module to obtain a target segmentation model, which is more suitable for the requirements of fine-component semantic segmentation of leaves, and while maintaining lightweight computing, ensures the accuracy of semantic segmentation.

[0056] In some embodiments, based on the semantic segmentation result, obtaining the phenotypic traits of the target by using high-throughput phenotyping technology includes: Obtaining a normalized semantic segmentation result by rotating and scaling the semantic segmentation result; Extracting basic phenotypic parameters of the target from the normalized semantic segmentation result by using high-throughput phenotyping technology; Quantifying the phenotypic traits of the target according to the basic phenotypic parameters.

[0057] Specifically, the embodiments of the present application first perform normalization processing on the semantic segmentation result, and then combine high-throughput phenotyping technology to obtain the phenotypic traits of the target.

[0058] For example, the semantic segmentation result is normalized by rotation and scaling to obtain a "dimensionless" image of lettuce leaves. The size of the lettuce leaf picture can be set to 1080×1080 to achieve a balance between computational efficiency and accuracy. A high-throughput lettuce leaf phenotypic trait extraction pipeline is constructed to extract basic phenotypic parameters of lettuce leaves (such as leaf length, leaf width, leaf perimeter, leaf area, etc.).

[0059] After obtaining the basic phenotypic parameters, the phenotypic traits of lettuce leaves can be quantified according to the UPOV (International Union for the Protection of New Varieties of Plants) guidelines, including shape traits such as leaf shape, leaf tip shape, leaf margin shape, and vein shape, and color traits such as leaf hue, color lightness, and anthocyanin content.

[0060] Leaf shape: The radial distance method is proposed. Assuming that the ideal shape of lettuce leaves is circular, the Euclidean distance from the center of the leaf to each point on the leaf edge is approximately equal. Therefore, the leaf is evenly divided into equal parts , and the Euclidean distance from the center of the leaf to each segmentation point is calculated respectively to score the approximate possibility of the leaf shape. The specific steps are as follows: First, perform circularity judgment, calculate the leaf length-width ratio according to the concept of leaf - class , if , locate the center point of the leaf , from to each segmentation point on the leaf edge calculate the radial distance , thus obtaining a radial distance , the formula is as follows (1) wherein, represents the coordinates of the blade center , represents the th segmentation point , represents the Euclidean distance from the blade center to the th segmentation point. Then, calculate the standard deviation of the radial distance , to measure the degree of the blade approaching a circular shape. The formula is: (2) (3) wherein, is the average value of the radial distance. The closer the value is to 1, the closer the blade shape is to a circular shape. According to the threshold division, horizontally broad oval , circular .

[0061] Then, judge the oval shape. If , according to the threshold division, broad oval , medium oval , narrow oval .

[0062] Secondly, judge the linear shape and lanceolate shape. If , select the blade vertex and the 50th points adjacent to the left and right to form an included angle, and calculate the included angle . According to the threshold division, lanceolate , linear .

[0063] Meanwhile, in each judgment step, the judgment of rhombus and triangle is also carried out according to the edge relation theorem. Finally, according to the UPOV guidelines, the blade shapes are divided into horizontally narrow circular, circular, horizontally broad oval, broad oval, medium oval, narrow oval, linear, lanceolate, rhombus, and triangle.

[0064] Leaf tip shape: The leaf tip shape mainly refers to the shape at the top of the blade. Extract the top 1 / 5 part of the blade as the region of interest, and calculate the angles of the concave and convex contours by approximating the polygon contour. Let the coordinates of the leaf tip vertex be , the coordinates of the 50th point adjacent to the left be , and the coordinates of the 50th point adjacent to the right be , then the included angle​​​​​​​​​​​​​​​​​ The calculation formula is as follows: (4) Among them , in order to more intuitively reflect the concavity and convexity of the blade, the angle is converted to , and the formula is as follows: (5) According to the angle difference, the leaf tip shape is divided into sharp tip , blunt tip , round , reverse heart and other shapes.

[0065] Leaf margin shape: First, extract the edge contour coordinate points, and use the interval sampling method to calculate the included angle between adjacent points (the 10th coordinate points adjacent to the left and right) in a loop. The method is the same as formulas (4) and (5). Taking the straight angle as the demarcation point, if it is less than 180 degrees, it is concave inward, and the farther the distance, the more severe the concavity. Similarly, if it is greater than 180 degrees, it is convex outward, and the farther the distance, the more severe the convexity. According to the severity (taking as a grading) and the proportion, the leaf margin is divided into entire margin, serrated, obtuse scale, double serrated, etc.

[0066] Leaf vein shape: At the mature stage, the leaf vein structure is complex and difficult to extract, and it is difficult to locate the included angle between leaf veins. The Hough line transform is used to detect the leaf vein lines. In the Cartesian coordinate system, the equation of a straight line is , where is a certain point on the leaf vein (i.e., the edge point), is the slope, is the intercept. In order to use the polar coordinate representation of the straight line in the Hough transform, the straight line equation is converted into the polar coordinate form , where is the distance from the straight line to the origin, is the angle of the normal line of the straight line. First, for each edge point select different values to calculate the corresponding , then, in the parameter space record the voting number of each pair of , and through the value of the maximum point, obtain the angle transformation formula of the straight line: (6) Finally, through mean normalization and division according to the threshold, the leaf vein shape is divided into fan-shaped , semi-fan-shaped and non-fan-shaped .

[0067] Leaf hue, color lightness, anthocyanin content: Color directly reflects the chlorophyll content, anthocyanin content, photosynthesis ability, and environmental adaptability of leaves. To more intuitively reflect hue and lightness, the RGB image is converted to the HSV color space. By averaging the H, S, and V components and setting thresholds, the leaf color is divided into yellow-green, green, and gray-green, and the color lightness is divided into very dark, dark, medium, bright, and very bright, as well as the degree of anthocyanin coloring, etc.

[0068] Among them, phenotypic traits may include the target morphology, agronomic traits, and all measurable physical characteristics, etc.

[0069] The high-throughput DUS testing method provided by the embodiments of the present application realizes the normalization of the semantic segmentation results by rotating and scaling the semantic segmentation results, avoids the influence of different sizes of the semantic segmentation results on trait acquisition, improves the accuracy of phenotypic trait acquisition, and can improve the efficiency of phenotypic analysis.

[0070] In some embodiments, determining the variety of the crop to be identified based on the phenotypic traits includes: Obtaining variety fingerprints of different varieties based on the phenotypic traits of different varieties; Determining the variety of the crop to be identified based on the variety fingerprint and the image of the crop to be identified.

[0071] Specifically, variety fingerprints of different varieties are constructed based on the phenotypic traits of different varieties.

[0072] For example, use statistical methods to analyze the basic phenotypic parameters of lettuce leaves, calculate indicators such as the average value and standard deviation, and exclude extreme outliers. Then, standardize the data to eliminate the influence of different dimensions on the analysis. Finally, combine the phenotypic parameters into a comprehensive trait index. The comprehensive trait index corresponding to each variety characterizes the phenotypic traits of the variety, and a radar chart can be generated according to the comprehensive trait index corresponding to each variety, so as to obtain unique phenotypic fingerprints (i.e., variety fingerprints) of multiple lettuce varieties.

[0073] The high-throughput DUS testing method provided by the embodiments of the present application generates unique digital fingerprint information for each variety, which can provide a direct and effective tool for variety identification and management, realize fast and accurate variety identification and traceability. Especially in the case where molecular detection is inconvenient, relatively accurate variety differentiation can also be achieved through variety fingerprints, which helps in the registration and protection of new varieties and promotes the management of intellectual property rights.

[0074] In some embodiments, determining the variety of the crop to be identified based on the variety fingerprint and the image of the crop to be identified includes: Obtaining the phenotypic traits of the crop to be identified based on the image of the crop to be identified; The variety of the crop to be identified is determined by comparing the phenotypic traits of the crop to be identified with the variety fingerprints.

[0075] Specifically, in the actual application process, for the crop to be identified, an image of the crop to be identified is collected, the phenotypic traits of the crop to be identified are obtained based on the image, and then the phenotypic traits of the crop to be identified are compared with the pre-constructed variety fingerprints of each variety, so as to determine the variety of the crop to be identified.

[0076] The high-throughput DUS testing method provided by the embodiments of the present application, by introducing high-throughput image acquisition and analysis technologies, combining machine learning and deep learning algorithms, establishes an automated DUS test trait identification pipeline, which can automatically and batchwise obtain multi-dimensional phenotypic data of the target, greatly improve the speed and efficiency of DUS testing, save human resources, help breeders and research institutions accelerate the breeding process of new varieties, and promote the protection and utilization of lettuce variety diversity; using computer vision and data analysis technologies, reduce the subjective errors brought by manual measurement, and ensure the accuracy and consistency of DUS test results; through in-depth analysis of phenotypic data, screen out key trait features, and establish standardized variety fingerprints, providing a scientific basis for the identification of new varieties and variety protection. In addition, the embodiments of the present application can quickly identify variety characteristics, and can be extended to phenotypic analysis, variety identification and breeding research of various crops, with high application value and promotion potential, providing support for breeding research and new variety registration, and helping to promote innovation and development.

[0077] Figure 2 is a schematic structural diagram of the high-throughput DUS testing device provided by the present invention, as Figure 2 shown, the present invention provides a high-throughput DUS testing device, including a first acquisition module 201, a second acquisition module 202 and a determination module 203.

[0078] The first acquisition module 201 is used to perform semantic segmentation on the to-be-processed image of the target based on the target segmentation model, and obtain a semantic segmentation result; the target segmentation model is a semantic segmentation model constructed based on the DenseASPP (Dense Connected Atrous Spatial Pyramid Pooling) module and the MSDA (Multi-Scale Dilated Attention) module; The second acquisition module 202 is used to obtain the phenotypic traits of the target by using high-throughput phenotypic technology based on the semantic segmentation result; The determination module 203 is used to determine the variety of the crop to be identified based on the phenotypic traits.

[0079] In some embodiments, the determination module 203 includes: The first acquisition unit is used to obtain the variety fingerprints of different varieties based on the phenotypic traits of different varieties; A first determination unit, configured to determine the variety of the crop to be identified based on the variety fingerprint and the image of the crop to be identified.

[0080] In some embodiments, the first determination unit includes: An acquisition subunit, configured to acquire the phenotypic traits of the crop to be identified based on the image of the crop to be identified; A determination subunit, configured to determine the variety of the crop to be identified by comparing the phenotypic traits of the crop to be identified with the variety fingerprint.

[0081] In some embodiments, the second acquisition module 202 includes: A second acquisition unit, configured to obtain a normalized semantic segmentation result by rotating and scaling the semantic segmentation result; A third acquisition unit, configured to extract the basic phenotypic parameters of the target from the normalized semantic segmentation result by using high-throughput phenotyping technology; A fourth acquisition unit, configured to quantify the phenotypic traits of the target according to the basic phenotypic parameters.

[0082] In some embodiments, the target segmentation model is a semantic segmentation model DeepLabV3+ improved based on the DenseASPP module and the MSDA module, and the backbone network of the improved DeepLabV3+ is a lightweight convolutional neural network ShuffleNetV2.

[0083] In some embodiments, it further includes: An acquisition module, configured to acquire target images of different varieties of crops; A third acquisition module, configured to perform data annotation and data augmentation on the target image to obtain the image to be processed of the target.

[0084] Specifically, the above-mentioned high-throughput DUS testing device provided by the present invention can implement all the method steps implemented by the above-mentioned high-throughput DUS testing method embodiments, and can achieve the same technical effects. The same parts and beneficial effects as those in the method embodiments in this embodiment will not be specifically described herein.

[0085] It should be noted that the division of units / modules in the above-mentioned embodiments of the present invention is illustrative, only a logical function division, and there may be other division methods in actual implementation. In addition, in each embodiment of the present application, each functional unit may be integrated in one processing unit, or each unit may exist physically alone, or two or more units may be integrated in one unit. The above-mentioned integrated units may be implemented in the form of hardware or in the form of software functional units.

[0086] Figure 3This is a schematic structural diagram of the electronic device provided by the present invention. As Figure 3 shown, the electronic device may include: a processor 301, a communications interface 302, a memory 303, and a communication bus 304. Among them, the processor 301, the communications interface 302, and the memory 303 complete mutual communication through the communication bus 304. The processor 301 may call the logical instructions in the memory 303 to execute the high-throughput DUS test method, and the method includes: Performing semantic segmentation on the to-be-processed image of the target based on the target segmentation model to obtain a semantic segmentation result; the target segmentation model is a semantic segmentation model constructed based on the DenseASPP (Dense Connected Atrous Spatial Pyramid Pooling) module and the MSDA (Multi-Scale Dilated Attention) module; Obtaining the phenotypic traits of the target by using high-throughput phenotyping technology based on the semantic segmentation result; Determining the variety of the crop to be identified based on the phenotypic traits.

[0087] In some embodiments, the determining the variety of the crop to be identified based on the phenotypic traits includes: Obtaining the variety fingerprints of different varieties based on the phenotypic traits of different varieties; Determining the variety of the crop to be identified based on the variety fingerprints and the image of the crop to be identified.

[0088] In some embodiments, the determining the variety of the crop to be identified based on the variety fingerprints and the image of the crop to be identified includes: Obtaining the phenotypic traits of the crop to be identified based on the image of the crop to be identified; Determining the variety of the crop to be identified by comparing the phenotypic traits of the crop to be identified with the variety fingerprints.

[0089] In some embodiments, the obtaining the phenotypic traits of the target by using high-throughput phenotyping technology based on the semantic segmentation result includes: Obtaining a normalized semantic segmentation result by rotating and scaling the semantic segmentation result; Extracting the basic phenotypic parameters of the target from the normalized semantic segmentation result by using high-throughput phenotyping technology; Quantifying the phenotypic traits of the target according to the basic phenotypic parameters.

[0090] In some embodiments, the target segmentation model is a semantic segmentation model DeepLabV3+ improved based on the DenseASPP module and the MSDA module, and the backbone network of the improved DeepLabV3+ is a lightweight convolutional neural network ShuffleNetV2.

[0091] In some embodiments, the method further includes: Collecting target images of different varieties of crops; Performing data annotation and data augmentation on the target images to obtain the to-be-processed images of the targets.

[0092] Specifically, the processor 301 may be a central processing unit (CPU), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA), or a complex programmable logic device (CPLD). The processor may also adopt a multi-core architecture.

[0093] When the logical instructions in the memory 303 are implemented in the form of software functional units and sold or used as independent products, they may be stored in a processor-readable storage medium. Based on such an understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, may be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) or a processor to execute all or part of the steps of the methods described in the various embodiments of the present application. The foregoing storage medium includes: various media such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disc that can store program codes.

[0094] In some embodiments, a computer program product is further provided. The computer program product includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the high-throughput DUS test method provided in the foregoing method embodiments. The method includes: Perform semantic segmentation on the image to be processed of the target based on the target segmentation model, and obtain a semantic segmentation result; the target segmentation model is a semantic segmentation model constructed based on the Dense Connected Atrous Spatial Pyramid Pooling (DenseASPP) module and the Multi-Scale Dilated Attention (MSDA) module; Obtain the phenotypic traits of the target by using the high-throughput phenotyping technology based on the semantic segmentation result; Determine the variety of the crop to be identified based on the phenotypic traits.

[0095] Specifically, the above computer program product provided by the embodiments of the present application can implement all the method steps implemented by the above method embodiments, and can achieve the same technical effects. The same parts and beneficial effects as those in the method embodiments will not be specifically described herein.

[0096] In some embodiments, a computer-readable storage medium is further provided. The computer-readable storage medium stores a computer program, and the computer program is used to cause a computer to execute the high-throughput DUS testing method provided by the above method embodiments. The method includes: Perform semantic segmentation on the image to be processed of the target based on the target segmentation model, and obtain a semantic segmentation result; the target segmentation model is a semantic segmentation model constructed based on the Dense Connected Atrous Spatial Pyramid Pooling (DenseASPP) module and the Multi-Scale Dilated Attention (MSDA) module; Obtain the phenotypic traits of the target by using the high-throughput phenotyping technology based on the semantic segmentation result; Determine the variety of the crop to be identified based on the phenotypic traits.

[0097] Specifically, the above computer-readable storage medium provided by the present invention can implement all the method steps implemented by the above method embodiments, and can achieve the same technical effects. The same parts and beneficial effects as those in the method embodiments will not be specifically described herein.

[0098] It should be noted that the computer-readable storage medium can be any available medium or data storage device accessible by the processor, including but not limited to magnetic memories (such as floppy disks, hard disks, magnetic tapes, magneto-optical discs (MO), etc.), optical memories (such as CDs, DVDs, BDs, HVDs, etc.), and semiconductor memories (such as ROM, EPROM, EEPROM, non-volatile memories (NAND FLASH), solid-state drives (SSD)).

[0099] It should be further noted that in the present invention, terms such as "first" and "second" are used to distinguish similar objects, rather than to describe a specific order or sequence. It should be understood that such terms can be interchanged under appropriate circumstances, so that the embodiments of the present invention can be implemented in an order other than those illustrated or described herein, and the objects distinguished by "first" and "second" are generally of the same type, and the number of objects is not limited. For example, the first object can be one or more.

[0100] "Determining B based on A" in the present invention means that the factor A should be considered when determining B. It is not limited to "determining B only based on A", but also includes: "determining B based on A and C", "determining B based on A, C, and E", "determining C based on A and further determining B based on C", etc. In addition, it can also include using A as a condition for determining B. For example, "when A meets the first condition, use the first method to determine B"; for another example, "when A meets the second condition, determine B"; for another example, "when A meets the third condition, determine B based on the first parameter", etc. Of course, it can also be a condition for using A as a factor for determining B. For example, "when A meets the first condition, use the first method to determine C and further determine B based on C", etc.

[0101] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage and optical storage, etc.) containing computer-usable program code.

[0102] The present application is described with reference to the flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or block in the flowchart and / or block diagram, and the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer-executable instructions. These computer-executable instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate means for implementing the functions specified in one Figure 1 one process or multiple processes and / or blocks Figure 1 one block or multiple blocks.

[0103] These processor-executable instructions may also be stored in a processor-readable memory that can direct a computer or other programmable data processing apparatus to operate in a particular manner, such that the instructions stored in the processor-readable memory produce a manufacture including an instruction means that implements the functions specified in one process Figure 1 or a plurality of processes and / or blocks Figure 1 or a plurality of blocks.

[0104] These processor-executable instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, whereby the instructions executed on the computer or other programmable apparatus provide steps for implementing the functions specified in one process Figure 1 or a plurality of processes and / or blocks Figure 1 or a plurality of blocks.

[0105] It will be apparent to those skilled in the art that various modifications and variations can be made to the present invention without departing from the spirit and scope of the invention. Thus, if these modifications and variations of the present invention come within the scope of the claims of the present invention and their equivalent technologies, the present invention is also intended to include these modifications and variations.

Claims

1. A high-throughput DUS testing method, characterized in that: include: Based on the target segmentation model, semantic segmentation is performed on the target image to be processed to obtain a semantic segmentation result; the target segmentation model is a semantic segmentation model constructed based on a densely connected void space pyramid pooling DenseASPP module and a multi-scale expansion attention MSDA module; Based on the semantic segmentation result, high-throughput phenotyping technology is used to obtain the phenotypic traits of the target; The variety of the crop to be identified is determined based on the phenotypic traits.

2. The high-throughput DUS testing method according to claim 1, characterized in that: The step of determining the variety of the crop to be identified based on the phenotypic traits comprises: Obtaining variety fingerprints of different varieties based on their phenotypic traits; The variety of the crop to be identified is determined based on the variety fingerprint and the image of the crop to be identified.

3. The high-throughput DUS testing method according to claim 2, characterized in that: The step of determining the variety of the crop to be identified based on the variety fingerprint and the image of the crop to be identified includes: Acquiring phenotypic traits of the crop to be identified based on the image of the crop to be identified; The variety of the crop to be identified is determined by comparing the phenotypic traits of the crop to be identified with the variety fingerprint.

4. The high-throughput DUS testing method according to claim 1, characterized in that: The method of obtaining the phenotypic traits of the target by using high-throughput phenotyping technology based on the semantic segmentation result includes: Obtaining a normalized semantic segmentation result by rotating and scaling the semantic segmentation result; Extracting basic phenotypic parameters of the target from the normalized semantic segmentation results using high-throughput phenotyping technology; The phenotypic traits of the target are quantified according to the basic phenotypic parameters.

5. The high-throughput DUS testing method according to claim 1, characterized in that: The target segmentation model is an improved semantic segmentation model DeepLabV3+ based on the DenseASPP module and the MSDA module, and the backbone network of the improved DeepLabV3+ is a ShuffleNetV2 network constructed by lightweight deep separable convolution and channel rearrangement operations.

6. The high-throughput DUS testing method according to claim 1, characterized in that: The method further comprises: Collect target images of different varieties of crops; Data annotation and data expansion are performed on the target image to obtain an image to be processed of the target.

7. A high-throughput DUS testing device, characterized in that: include: The first acquisition module is used to perform semantic segmentation on the target image to be processed based on the target segmentation model to obtain a semantic segmentation result; the target segmentation model is a semantic segmentation model constructed based on a densely connected void space pyramid pooling DenseASPP module and a multi-scale expansion attention MSDA module; A second acquisition module is used to acquire the phenotypic traits of the target using high-throughput phenotyping technology based on the semantic segmentation result; A determination module is used to determine the variety of the crop to be identified based on the phenotypic traits.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the high-throughput DUS testing method according to any one of claims 1 to 6 is implemented.

9. A non-transitory computer-readable storage medium, characterized in that: The non-transitory computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the high-throughput DUS testing method according to any one of claims 1 to 6 is implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the high-throughput DUS testing method according to any one of claims 1 to 6 is implemented.

Images