A method and system for intelligently detecting the phenotype of a picea abies mature cotyledon embryo and a storage medium

Abstract

The application relates to a European spruce mature cotyledon embryo phenotype intelligent detection method and system and a storage medium, and relates to the cross technical field of forest tree breeding and computer vision. The detection method in the application first constructs a mobile phone image data set of European spruce somatic embryo differentiation and performs boundary box labeling; secondly, a MultiSeamHead module is introduced based on a YOLOv11n framework, the detection capability for small targets, dense occlusion and complex backgrounds is enhanced through multi-scale feature fusion and a spatial attention mechanism; then, the improved detection model is trained by using the labeled data set; finally, a visual interactive platform is constructed based on PySide6, the trained model is integrated, and the functions of image import, automatic detection, result visualization and data export are realized. Experiments show that the precision of the improved model reaches 0.861, the recall rate reaches 0.863, and the mAP@50 reaches 0.927, which is significantly improved compared with the original model, efficient, accurate and high-throughput mature cotyledon embryo phenotype automatic detection is realized, and the technical application threshold is reduced.

Description

Technical Field

[0001] This application relates to the interdisciplinary field of forest tree breeding and computer vision, and in particular to a method, system and storage medium for intelligent detection of mature cotyledon embryo phenotype in European spruce. Background Technology

[0002] European spruce ( Picea abies (L.) Karst., as one of the most important timber species in Northern and Central Europe, relies on somatic embryogenesis technology for rapid clonal propagation and genetic improvement of superior forest tree varieties. The somatic embryogenesis process of conifers generally exhibits significant asynchrony; that is, in the same culture system, multiple developmental stages of somatic embryos, including early somatic embryos, mid-stage somatic embryos, early cotyledonary embryos, mid-cotyledonary embryos, and mature cotyledonary embryos, often coexist. Only mature cotyledonary embryos can develop into somatic seedlings through drying and germination treatments. Therefore, the number of differentiated mature cotyledonary embryos and their geometric morphology (e.g., length and width) are important phenotypic parameters for evaluating the quality of embryogenic cell lines and predicting seedling potential. Accurate and high-throughput acquisition of these parameters is of great significance for accelerating forest tree breeding and promoting industrialized seedling production.

[0003] Currently, the counting of somatic embryos mainly relies on researchers observing and counting under a microscope or with the naked eye. This method is not only labor-intensive and time-consuming, but also prone to observational errors due to operator subjectivity, resulting in poor repeatability and making it difficult to meet the needs of large-scale, high-throughput, and precise analysis. Therefore, there is an urgent need to develop a method that can automatically, quickly, and accurately identify the number of mature cotyledonary embryos and measure their geometric parameters.

[0004] In recent years, computer vision technology centered on deep learning has provided a new solution for achieving objective and high-throughput evaluation of target features. Among them, the YOLO (You Only Look Once) series of algorithms has been widely used in industrial quality inspection, remote sensing image analysis, and other fields due to its end-to-end detection speed and superior accuracy. YOLOv11, the latest version released by Ultralytics, significantly improves accuracy and robustness in complex scenes and small target detection by introducing the C2PSA (Parallel Spatial Attention) module. In particular, its lightweight version, YOLOv11n, has the advantages of small parameter size and fast inference speed, making it suitable for deployment on mobile or embedded devices.

[0005] However, directly applying YOLOv11n to the detection of mature cotyledon embryos in European spruce still faces serious challenges. European spruce somatic embryo images have the following characteristics: (1) The target size is extremely small: The actual pixel length and width of mature cotyledon embryos are mainly distributed between 20 and 50 pixels. According to the definition of small targets (area less than 32×32 pixels) in the COCO dataset, they are typical micro-targets; (2) Densely distributed and mutually occluded: Somatic embryos often show dense clustering or even mutual occlusion in the culture system, which brings great difficulties to bounding box regression and individual counting; (3) Complex background and low contrast: There are often a large number of non-embryonic tissues around somatic embryos, with complex background and low contrast, which easily leads to missed detection and false detection.

[0006] Therefore, directly applying the YOLOv11n model is insufficient to achieve the desired detection accuracy. There is an urgent need to make targeted improvements to it and develop supporting visualization tools to lower the technical application threshold and achieve high-throughput automated detection. Summary of the Invention

[0007] The purpose of this invention is to overcome the problems of low efficiency, high subjectivity, and poor repeatability in the existing technology that relies on manual counting of embryos and measurement of geometric parameters. It provides an intelligent detection method, system and storage medium for mature cotyledon embryo phenotype of European spruce. By improving the target detection model and customizing the graphical user interface, it enables efficient and accurate identification of the number of mature cotyledon embryos and measurement of embryo length and width by taking pictures with ordinary mobile phones.

[0008] To achieve the above objectives, the present invention provides the following technical solution: A method for intelligent detection of mature cotyledon embryo phenotype in European spruce includes the following steps: Step 1: Obtain original mobile phone images of somatic embryo differentiation of European spruce and construct a somatic embryo phenotype dataset; Step 2: Perform image annotation and dataset partitioning on the somatic embryo phenotype dataset. The image annotation includes bounding box information of the number of mature cotyledon embryos, somatic embryo length, and somatic embryo width. The dataset is divided into training set, validation set, and test set. Step 3: Construct an improved YOLOv11n detection model. The improvements include: introducing the MultiSeamHead module based on the YOLOv11n model. The MultiSeamHead module is used to improve the detection performance of small targets in occluded scenes to adapt to the detection of mature cotyledon embryo phenotypes of European spruce. Step 4: Train the improved YOLOv11n detection model using the labeled dataset to obtain the trained somatic embryo phenotype detection model; Step 5: Build a visual interactive platform based on PySide6, and deploy the trained somatic embryo phenotypic detection model on the platform. The platform is used to receive mobile phone images uploaded by users, call the model for inference in real time, and visualize the detection results. The detection results include at least the number of mature cotyledon embryos, somatic embryo length, and somatic embryo width.

[0009] Furthermore, in step one, 10 embryogenic cell lines of European spruce with different growth characteristics were used as experimental materials. The original images of somatic embryo differentiation of European spruce were acquired by mobile phone using a fixed-distance photography method. During the acquisition process, optical zoom variables were introduced. The image resolution included two specifications: 3072×3072 and 1024×1024. The magnification included three optical zoom levels: 1X, 1.2X, and 2.1X. Finally, a somatic embryo phenotype dataset containing 10 cell lines, 2 image sizes, and 3 magnifications was constructed.

[0010] Furthermore, step two, which involves labeling the images and partitioning the dataset, includes: Based on the classical morphological definition of a mature cotyledon embryo, namely, a fully developed cotyledon, a clear hypocotyl, and a plump shape, all images in the somatic embryo phenotype dataset were manually labeled using Labelme image annotation software. The original images of the collected European spruce somatic embryos were imported into Labelme, and the YOLO annotation mode was selected. The target region of the mature cotyledon embryo in the image was selected by manually dragging a rectangle and named ME2. After each annotation is completed, a JSON annotation file with the same name as the image is generated in the current image directory, recording the location information of all mature cotyledon embryos in the image; the annotation file adopts the YOLO standard format and includes the target category name, the normalized coordinates of the target bounding box center point, and the width and height of the target bounding box; After annotation, all images and their corresponding annotation files are divided into training, validation, and test sets in a 7:2:1 ratio to provide data for subsequent model training and performance evaluation.

[0011] Furthermore, the MultiSeamHead module replaces the original detection head and enhances the target representation capability through multi-scale feature fusion, spatial attention enhancement, and global context information extraction, specifically including: (1) Enhance the expression of tiny features: The pixel size of mature cotyledon embryos of European spruce is mainly distributed between 20 and 50 pixels, which is a typical tiny target. The MultiSeamHead module decouples the features of classification tasks and regression tasks through a multi-branch semantic decoupling strategy, which preserves and enhances high-resolution and detailed semantic information for tiny targets, and avoids the loss or submersion of features by the background during the downsampling process of deep network, thereby improving the recall rate of tiny embryos; (2) Improve occlusion discrimination capability: In response to the problem of feature overlap caused by the clustering and mutual occlusion of somatic embryos, the MultiSeamHead module adopts a context aggregation strategy, combines the surrounding environment information of the target to infer the complete target, strengthens the feature response to the occluded part, and improves the positioning accuracy and individual separation capability in densely arranged scenes. (3) Enhanced feature recognition: In response to background interference caused by non-embryonic tissues and somatic embryos at other developmental stages, the MultiSeamHead module introduces a spatial attention mechanism to concentrate computational resources on the key features of mature cotyledon embryos, while suppressing redundant background information, significantly improving the distinction between the target and the background and reducing the false detection rate.

[0012] Furthermore, the visualization and interaction platform is developed based on the PySide6 framework, providing a graphical user interface, and its main functional modules include: Image import module: Supports importing single images of embryonic bodies taken with a mobile phone, compatible with common image formats; Model calling module: Loads the trained improved YOLOv11n model and performs automatic detection on the imported images; Results display module: Displays the detected image in the graphical user interface, marks the bounding box of each mature cotyledon embryo, and counts the total number of mature cotyledon embryos in real time; Data export module: Exports the detected number of mature cotyledon embryos, somatic embryo length, and somatic embryo width data into a CSV file and generates a visualization image containing the detection results.

[0013] This invention also provides an intelligent detection system for the mature cotyledon embryo phenotype of European spruce, comprising: Image acquisition module: used to capture raw images of somatic embryo differentiation of European spruce using a mobile phone; Data processing module: used to annotate bounding boxes on the original images and build the training dataset; Model building module: used to build the improved YOLOv11n detection model as described above; Model training module: Used for training and validating the improved YOLOv11n detection model; Visualization and Interaction Module: Developed based on the PySide6 framework, it provides a graphical user interface, integrates pre-trained models, and enables functions such as image import, real-time detection, result visualization, and data export.

[0014] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described intelligent detection method for mature cotyledon embryo phenotype of European spruce.

[0015] The present invention also provides a computer program product, including a computer program or instructions, which, when executed by a processor, implement the steps of the above-described intelligent detection method for mature cotyledon embryo phenotype of European spruce.

[0016] Compared with the prior art, the present invention has the following beneficial effects: 1. Improved detection efficiency and objectivity: This invention achieves automated identification and counting of mature cotyledon embryos of European spruce by constructing an improved YOLOv11n target detection model, replacing the traditional manual microscope observation method, significantly improving detection efficiency, eliminating observation errors caused by human subjective factors, and ensuring the objectivity and repeatability of detection results.

[0017] 2. Improved detection performance for small targets and occluded scenes: This invention introduces the MultiSeamHead module into the YOLOv11n detection head. Through multi-scale feature fusion, spatial attention enhancement, and context aggregation strategies, it effectively solves the detection challenges of small target size, dense distribution, mutual occlusion, and complex backgrounds in European spruce embryo images. Experimental results show that the improved model achieves a precision of 0.861, a recall of 0.863, and an mAP@50 of 0.927, representing improvements of 2.1%, 2.4%, and 2.7% respectively compared to the native YOLOv11n, while reducing computational cost, achieving a better balance between accuracy and efficiency.

[0018] 3. Lowering the technical application threshold: This invention develops a visual interactive platform based on PySide6, which integrates a pre-trained detection model. Users only need to take pictures of the somatic embryos with a regular mobile phone and upload them to the platform to automatically obtain key phenotypic parameters such as the number of mature cotyledon embryos and the length and width of the somatic embryos. No professional computer vision background knowledge is required, which facilitates its promotion and application in forest tree breeding research and industrial seedling production scenarios.

[0019] 4. Supports high-throughput data acquisition and analysis: Through lightweight model design and automated platform processing, this invention supports high-throughput detection of large-scale somatic embryo images. The detection results can be automatically exported as structured data, providing reliable data support for subsequent statistical analysis, cell line quality evaluation and breeding of superior varieties. Attached Figure Description

[0020] The above and other objects, features and advantages of this application will become more apparent from the more detailed description of exemplary embodiments thereof in conjunction with the accompanying drawings, wherein the same reference numerals generally represent the same components in the exemplary embodiments thereof.

[0021] Figure 1 This is an overall flowchart of the intelligent detection method for mature cotyledon embryo phenotype of European spruce in this embodiment of the invention; Figure 2This is a schematic diagram of the network structure of the improved YOLOv11n model in an embodiment of the present invention, in which the specific location and internal structure of the MultiSeamHead detection head are marked; Figure 3 This is a schematic diagram illustrating the training process of the improved YOLOv11n model loss function in an embodiment of the present invention; Figure 4 This is a schematic diagram of the confusion matrix of the detection results of the improved YOLOv11n model in an embodiment of the present invention; Figure 5 This is a comparison chart of the performance of the YOLOv11n model and the YOLOv11n-MultiSeamHead model (improved model) in the mature cotyledon embryo detection task in this embodiment of the invention; Figure 6 This is a schematic diagram of a visual interactive platform based on PySide6 in an embodiment of the present invention; wherein, Figure 6 A in the middle represents the login interface of the visual interactive platform. B stands for: the homepage of the visual interactive platform; C stands for: Image detection interface of the visual interactive platform; D stands for: The image detection results interface of the visual interactive platform; E represents the resulting image (annotated image containing confidence threshold information). F stands for: Result image (including labeled image with somatic embryo number); G stands for: CSV result table. Detailed Implementation

[0022] Embodiments of this application will now be described in more detail with reference to the accompanying drawings. While embodiments of this application are shown in the drawings, it should be understood that this application may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to make this application more thorough and complete, and to fully convey the scope of this application to those skilled in the art.

[0023] Example This embodiment provides a method for intelligent detection of the somatic embryo phenotype of European spruce, the overall process of which is as follows: Figure 1 As shown, the specific steps include the following: Step 1: Dataset Construction and Preprocessing Using a mobile phone under supplemental lighting, images of culture dishes depicting somatic embryo differentiation of European spruce were taken at a fixed distance. Ten embryogenic cell lines of European spruce with different growth characteristics were used as experimental materials. Images were taken at two resolutions (3072×3072 and 1024×1024), and optical zoom was incorporated during the acquisition process, including three magnifications: 1X, 1.2X, and 2.1X, resulting in a total of 373 images.

[0024] Based on the classical morphological definition of a mature cotyledonous embryo (i.e., fully developed cotyledons, clear hypocotyl, and plump morphology), bounding boxes were annotated using the Labelme tool and named ME2. After each annotation, the software automatically generated a JSON file with the same name as the image in the current image directory, recording the category and location information of all targets in the image. The annotation file adopted YOLO format, recording in detail the name of each bounding box, the normalized coordinates of the bounding box center point, and the width and height of the box. After annotation, researchers with a background in somatic embryology performed double verification to reduce false positives and false negatives. The final dataset contains 8538 annotated instances. The dataset was randomly divided into training, validation, and test sets in a 7:2:1 ratio to provide high-quality data support for subsequent model training and performance evaluation.

[0025] Images of European spruce somatic embryos exhibit characteristics such as very small, densely distributed, and mutually occluded targets, along with complex and low-contrast backgrounds. According to dataset statistics, there are 8538 labeled instances, which is relatively few for a challenging target. To enhance the model's generalization ability, multi-level online augmentation was performed on the training set data during model training. Specific operations included: Mosaic enhancement: Four images are stitched together into one with 100% probability, and this enhancement strategy is turned off in the last 10 epochs, so that the model can learn the ability to locate objects in complex backgrounds and improve the robustness of detecting occluded objects.

[0026] Random flipping: The image is flipped horizontally with a 50% probability, which effectively increases the diversity of the data, allows the model to learn the mirror features of the target, and does not change the category of the target.

[0027] Color perturbation: Hue, saturation, and lightness are perturbed in the HSV color space with perturbation magnitudes of 0.2, 0.5, and 0.2, respectively, to simulate color changes caused by different lighting conditions and shooting devices, thereby enhancing the model's robustness to color changes. The larger saturation perturbation helps the model focus on shape features rather than color features, avoiding over-reliance on color information.

[0028] Geometric transformations include random ±30° rotation, random translation, 0-0.5x scaling, and shearing operations, which increase the diversity of samples in terms of scale, angle, and sharpness, and improve the robustness of the model to changes in viewpoint.

[0029] Random erasure: A rectangular region is randomly selected in the image and erased with a 30% probability to simulate the situation where the mature cotyledon embryo is partially occluded, thereby enhancing the model's ability to detect occluded targets.

[0030] Copy and paste enhancement: A detection target is randomly copied from other images in the same batch with a 30% probability, scaled and rotated, and then pasted onto the current image to increase the target density in the training image and significantly improve the detection performance of small targets.

[0031] Step 2: Improve YOLOv11n model construction This embodiment is an improvement based on the YOLOv11n framework. Figure 2 The diagram shows the network structure of the improved YOLOv11n model, with the specific location and internal structure of the MultiSeamHead detection head marked. The improvement lies in the introduction of the MultiSeamHead module into the detection head.

[0032] The MultiSeamHead detection head consists of two parts: a regression branch and a classification branch. The regression branch incorporates the MultiSeam module to enhance spatial feature modeling capabilities, thereby improving target localization accuracy. Specifically, the regression branch enhances feature representation through multi-scale convolution and adaptive pooling operations, and adaptively adjusts the importance of different features using a weight allocation mechanism, guiding the model to focus on key information. Subsequently, the Distribution Focal Loss (DFL) module transforms the discrete distribution into continuous bounding box coordinates. The classification branch first applies Depthwise Separable Convolution (DWConv) to the input features for efficient feature extraction, then increases the feature dimensionality through convolutional layers, further enhances the classification semantic information using the MultiSeam module, and finally maps the output to a multi-dimensional category through convolutional layers.

[0033] The MultiSeam module is a multi-head attention mechanism comprising three Channel and Spatial Mixing Modules (CSMMs) of different sizes, using patch sizes of 3, 5, and 7 respectively. The CSMM module effectively extracts multi-scale features through patches of different sizes, ensuring the network can handle information from multiple scales. Simultaneously, this module utilizes depthwise separable convolutions to learn the dependencies between spatial and channel dimensions, significantly reducing computational complexity while helping the network capture more nuanced spatial features and channel relationships. By comprehensively utilizing multi-scale features and depthwise separable convolutions, the CSMM module improves feature extraction accuracy while maintaining computational efficiency, making it particularly suitable for small object detection tasks.

[0034] By integrating MultiSeamHead into the YOLOv11n detection head, the model's ability to identify mature cotyledon embryo features in complex scenarios is effectively improved.

[0035] Step 3: Model Training and Experimental Environment Configuration All experiments in this embodiment were conducted on the Ubuntu operating system, using the PyTorch 2.5.1 deep learning framework to build, train, and test the YOLOv11n model. Model training was performed on an NVIDIA GeForce RTX 4090 GPU, with the input image size uniformly adjusted to 960×960 pixels and the batch size set to 10. The initial learning rate was set to 0.01, dynamically adjusted using a cosine annealing learning rate strategy, and the weight decay coefficient was set to 0.001. The number of training epochs was set to 1000, and an early stopping mechanism (patience=100) was introduced to prevent overfitting and save computational resources. The optimizer used stochastic gradient descent (SGD), and the loss function followed the original YOLOv11n loss function, including classification loss, regression loss, and confidence loss.

[0036] The experimental environment and parameter configurations are shown in Tables 1 and 2: Table 1 Hardware Environment Configuration Table 2 Software Environment Configuration Step 4: Model Evaluation and Validation The improved YOLOv11n model was trained using the processed somatic embryo differentiation image dataset, and the trained YOLOv11n model was used to detect mature cotyledon embryos of Picea pulcherrima.

[0037] Precision (P), recall (R), mAP@50, mAP@50-95, computational cost (GFLOPs), and parameters were used as evaluation metrics to validate the mature cotyledon detection performance of the YOLOv11n-MultiSeamHead model. The validation results are shown in Table 3.

[0038] Table 3 Comparative Ablation Experiments As shown in Table 3, after introducing MultiSeamHead, the model's precision, recall, mAP@50, and mAP@50-95 reached 0.861, 0.863, 0.927, and 0.605, respectively, representing improvements of 2.1%, 2.4%, 2.7%, and 0.9% compared to the native YOLOv11n. Notably, the YOLOv11n-MultiSeamHead model reduced its computational cost from 6.4 GFLOPs to 6.0 GFLOPs while improving accuracy, demonstrating superior computational efficiency. Although this model has slightly more parameters than the original model, it remains suitable for deployment in scenarios with high accuracy requirements and limited computational resources.

[0039] Figure 3 This is a schematic diagram illustrating the training process of the improved YOLOv11n model loss function in an embodiment of the present invention. Figure 3 As shown, during the training process, the box_loss, cls_loss, and dfl_loss of the model gradually decreased and tended to stabilize. The curves on the validation set and training set showed the same trend, indicating that the model did not exhibit significant overfitting. Regarding performance metrics, both precision and recall increased with the number of iterations, eventually stabilizing at approximately 0.9 and 0.6 for mAP@50 and mAP@50-95, respectively, validating the model's high precision and stability in the mature cotyledon embryo detection task of European spruce.

[0040] Figure 4 This is a schematic diagram of the confusion matrix of the detection results of the improved YOLOv11n model in an embodiment of the present invention. Figure 4 As shown in the confusion matrix results, the model in this application has good classification performance on both mature cotyledon embryos and background targets, with a low overall confusion level, indicating that the improved structure can effectively enhance the model's class discrimination ability.

[0041] Figure 5 This is a comparison chart showing the performance of the YOLOv11n model and the YOLOv11n-MultiSeamHead model (improved model) in the mature cotyledon embryo detection task in this embodiment of the invention. Figure 5 As shown, for simple detection targets (Figure A), the improved model is not significantly different from the original model in recognizing the number of mature cotyledon embryos, but the detection confidence is improved; for complex detection targets (Figure B), the improved model significantly improves the detection performance of occluded targets.

[0042] Step 5: Building a Visual Interactive Platform A visual interactive platform was developed based on the PySide6 framework. The platform interface is as follows: Figure 6 As shown.

[0043] The platform's main functions are as follows: User Login: Users can access the platform's main interface by entering their username and password.

[0044] Image Import: Users can select a single embryonic image taken with their mobile phone and import it via the "Upload Image" button.

[0045] Switching models: The platform loads the pre-trained weight file of the improved YOLOv11n model by default, but users can also manually select other pre-trained models to switch to.

[0046] Automatic detection: Click the "Start Detection" button, and the platform will call the current model to perform inference on the imported image. The detection progress will be displayed in real time.

[0047] Results Display: After the detection is completed, the right panel displays the labeled results with bounding boxes, the middle of the panel shows the location where the detection results are saved, and the lower left side shows the total number of mature cotyledon embryos in the current image.

[0048] Data export: The platform automatically exports the test results, including two images (an annotated image showing confidence threshold information and an annotated image showing the embryo number) and a CSV table (recording data such as the number of embryos, embryo length, and embryo width).

[0049] The various embodiments of this application have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen to best explain the principles, practical application, or improvement of the technology in the market, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. A method for intelligent detection of mature cotyledon embryo phenotype in European spruce, characterized in that, Includes the following steps: Step 1: Obtain original mobile phone images of somatic embryo differentiation of European spruce and construct a somatic embryo phenotype dataset; Step 2: Perform image annotation and dataset partitioning on the somatic embryo phenotype dataset. The image annotation includes bounding box information of the number of mature cotyledon embryos and the length and width of the somatic embryos. The dataset is divided into training set, validation set and test set. Step 3: Construct an improved YOLOv11n detection model. The improvements include: introducing the MultiSeamHead module based on the YOLOv11n model. The MultiSeamHead module is used to improve the detection performance of small targets in occluded scenes to adapt to the detection of mature cotyledon embryo phenotypes of European spruce. Step 4: Train the improved YOLOv11n detection model using the labeled dataset to obtain the trained somatic embryo phenotype detection model; Step 5: Build a visual interactive platform based on PySide6, and deploy the trained somatic embryo phenotypic detection model on the platform. The platform is used to receive mobile phone images uploaded by users, call the model for inference in real time, and visualize the detection results. The detection results include at least the number of mature cotyledon embryos, somatic embryo length, and somatic embryo width.

2. The intelligent detection method for mature cotyledon embryo phenotype of European spruce according to claim 1, characterized in that, In step one, 10 embryogenic cell lines of European spruce with different growth characteristics were used as experimental materials. The original images of somatic embryo differentiation of European spruce were acquired by mobile phone using a fixed-distance photography method. During the acquisition process, optical zoom variables were introduced. The image resolution included two specifications: 3072×3072 and 1024×1024. The magnification included three optical zoom levels: 1X, 1.2X and 2.1X. Finally, a somatic embryo phenotype dataset containing 10 cell lines, 2 image sizes and 3 magnifications was constructed.

3. The intelligent detection method for mature cotyledon embryo phenotype of European spruce according to claim 1, characterized in that, Step two involves image annotation and dataset partitioning of the somatic embryo phenotype dataset, including: Based on the classical morphological definition of a mature cotyledon embryo, namely, a fully developed cotyledon, a clear hypocotyl, and a plump shape, all images in the somatic embryo phenotype dataset were manually labeled using Labelme image annotation software. The original images of the collected European spruce somatic embryos were imported into Labelme, and the YOLO annotation mode was selected. The target region of the mature cotyledon embryo in the image was selected by manually dragging a rectangle and named ME2. After each annotation is completed, a JSON annotation file with the same name as the image is generated in the current image directory, recording the location information of all mature cotyledon embryos in the image; the annotation file adopts the YOLO standard format and includes the target category name, the normalized coordinates of the target bounding box center point, and the width and height of the target bounding box; After annotation, all images and their corresponding annotation files are divided into training, validation, and test sets in a 7:2:1 ratio for subsequent model training and performance evaluation.

4. The intelligent detection method for mature cotyledon embryo phenotype of European spruce according to claim 1, characterized in that, The MultiSeamHead module described in step three replaces the original detection head and includes a multi-scale feature fusion unit, a spatial attention enhancement unit, and a global context information extraction unit. It is used to enhance the model's ability to represent the features of small targets in order to reduce false detections and false negatives.

5. The intelligent detection method for mature cotyledon embryo phenotype of European spruce according to claim 1, characterized in that, The visualization and interaction platform described in step five is developed based on the PySide6 framework and includes the following functional modules: Image import module: Supports importing single images of embryonic bodies taken with a mobile phone; Model calling module: Loads the trained improved YOLOv11n model and performs automatic detection on the imported images; Results display module: Displays the detected image in the graphical user interface, marks the bounding box of each cotyledon embryo, and counts the total number of mature cotyledon embryos in real time; Data export module: Exports the detected number of embryos, embryo length, and embryo width as CSV files.

6. A smart detection system for the mature cotyledon embryo phenotype of European spruce, characterized in that, include: Image acquisition module: used to capture raw images of somatic embryo differentiation of European spruce using a mobile phone; Data processing module: used to annotate bounding boxes on the original images and build the training dataset; Model building module: used to build the improved YOLOv11n detection model as described in any one of claims 1-5; Model training module: Used for training and validating the improved YOLOv11n detection model; Visualization and Interaction Module: Developed based on the PySide6 framework, it provides a graphical user interface, integrates pre-trained models, and enables functions such as image import, real-time detection, result visualization, and data export.

7. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, which, when run on a computer, executes the intelligent detection method for mature cotyledon embryo phenotype of European spruce as described in any one of claims 1 to 5.

8. A computer program product, comprising a computer program or instructions, characterized in that, When the computer program or the instructions are executed by the computer, they implement the steps of the intelligent detection method for mature cotyledon embryo phenotype of European spruce as described in any one of claims 1 to 5.

Images