Method for identifying growing points of dicotyledonous plant seedlings based on image information

Abstract

The invention provides a method for identifying growing points of dicotyledonous plant seedlings based on image information. The method comprises the following steps: S1, optimizing a backbone network of a YOLOv8 model, and obtaining a YOLOv8-SEBlock-detect model and a YOLOv8-SEBlock-seg model; s2, a YOLOv8-SEBlock-detect model is trained, so that the YOLOv8-SEBlock-detect model can identify the growing points of the dicotyledonous plant seedlings; a YOLOv8-SEBlock-seg model is trained, so that the YOLOv8-SEBlock-seg model can identify the whole dicotyledonous plant seedling; s3, calling a YOLOv8-SEBlock-detect model to directly identify the growing points of all the dicotyledonous plant seedlings in the picture, and marking the growing points of all the dicotyledonous plant seedlings; and S4, calling a YOLOv8-SEBlock-seg model to identify all dicotyledonous plant seedlings in the picture, judging whether a growth point directly identified by the YOLOv8-SEBlock-detect model exists in the detection frame or not, if not, calculating a geometric center point of the detection frame, marking the geometric center point as a growth point, and replacing a real dicotyledonous plant seedling growth point position. According to the method, detection frame geometric center point replacement and model direct identification are combined, and the accuracy of dicotyledonous plant seedling growing point identification is improved.

Description

Technical Field

[0001] The present application relates to the field of smart agricultural technology, and more specifically, to a method for identifying the growth points of dicotyledonous plant seedlings based on image information. Background Art

[0002] Thinning dicot seedlings in agricultural production is typically done manually. Farmers visually select seedlings to be thinned and manually remove them. This process can easily damage the roots of other seedlings that don't need to be thinned. The process is also difficult, time-consuming, and labor-intensive, requiring careful, careful, and patient operators. Modern smart agricultural production technologies, such as automated thinning systems, have significantly promoted agricultural mechanization, reducing labor and improving efficiency.

[0003] The growth point is the part of a dicot seedling where cell division, proliferation, and growth are most active. It has a decisive influence on the morphological development of dicot seedlings. Identifying and damaging the growth point of a seedling can inhibit its growth and survival. Therefore, this phenomenon can be used to thin out and thin out seedlings in seedbed management.

[0004] Image recognition technology is an intelligent information processing technology based on computer vision and artificial intelligence. It simulates the perception mechanism of the human visual system and uses deep learning models to extract features, classify and locate target objects in images.

[0005] Traditional image recognition methods typically employ a two-stage detection framework. While highly accurate, this approach suffers from high computational complexity, making it difficult to meet real-time requirements. Extensive experiments have shown that the main errors in directly identifying growing points using target detection models are the inability to handle situations where the growing points of dicot seedlings are obscured or where the seedlings are too small to be identified. In practice, seedling holes typically contain two to three seedlings, and the inconsistent growth rates of dicot seedlings, resulting in small seedlings, and the obstruction of the growing point by leaves from other seedlings are common, severely impacting the automatic identification of dicot seedlings' growing points. Summary of the Invention

[0006] In view of this, the present application provides a method for identifying the growth points of dicotyledonous plant seedlings based on image information, which improves the accuracy of growth point detection by combining the replacement of the geometric center point of the detection frame with direct model recognition.

[0007] To achieve the above objectives, the technical solutions adopted in this application are as follows: The method for identifying the growth point of a dicotyledonous plant seedling based on image information comprises: S1: Optimize the YOLOv8 model backbone network. That is, embed the SEBlock (Squeeze-and-Excitation) unit after the C2f module of the YOLOv8 model backbone network to obtain the YOLOv8-SEBlock-detect model and the YOLOv8-SEBlock-seg model. S2: The YOLOv8-SEBlock-detect model is trained by manually annotating the dicot seedling growth point training set, so that it can identify the dicot seedling growth point; and the YOLOv8-SEBlock-seg model is trained by manually annotating the dicot seedling outline training set, so that it can identify the entire dicot seedling; S3: Load the two trained models simultaneously and call the YOLOv8-SEBlock-detect model to directly identify and mark the growth points of all dicot seedlings in the image. S4: Call the YOLOv8-SEBlock-seg model to identify all dicot seedlings in the image and store the dicot seedling external detection frame. Determine whether there is a growth point directly identified by the YOLOv8-SEBlock-detect model within the detection frame. If not, calculate the geometric center point of the detection frame and mark it as the growth point, replacing the actual dicot seedling growth point position.

[0008] Furthermore, the method further comprises: S5: Output and save the dicot seedling image after marking the growth point.

[0009] Furthermore, the method further comprises: S6: Based on images of dicot seedlings with marked growing points, monitor abnormal changes in the color or morphology of the growing points to assist in the detection of early lesions; or quantify the rate of change of the growing points to assist in stress resistance assessment; or assist in forest health monitoring and quantify the progress of ecological restoration; or combine with satellite imagery to track the changes in the growing points of dicot seedlings in alpine areas with altitude to study the impact of climate on the growth of dicot seedlings.

[0010] Furthermore, the dicotyledonous plant seedlings are tobacco seedlings, cotton seedlings, tomato seedlings, grape seedlings or forest tree seedlings.

[0011] Furthermore, the detection frame is a rectangular detection frame.

[0012] Compared with the prior art, the present invention has the following advantages: 1. Optimizing and improving the YOLOv8 model for specific agricultural scenarios, the SEBlock channel attention mechanism is embedded, enabling the model to dynamically learn the importance weights of different feature channels, automatically enhancing key features related to the growth points of dicot seedlings (such as edge texture and morphological contours) while suppressing redundant background information (such as soil noise and leaf occlusions). Compared to the traditional YOLOv8's equalized feature processing, the improved model significantly improves the discriminability of feature representation, achieving a 9.90% increase in mean average precision (mAP) on a self-built dataset.

[0013] 2. The model improvement steps in this application improve the accuracy of identifying the growth points of dicotyledonous seedlings, reduce the labor cost of naked eye detection, and lay a technical foundation for the realization of fully automatic machine thinning or transplanting.

[0014] 3. The improved YOLOv8-SEBlock-seg model significantly improves the accuracy of identifying dicot seedlings. This ensures that the detection frame is accurately drawn around the dicot seedlings and that the geometric center point of the detection frame can scientifically replace the actual growth point in most cases.

[0015] 4. In order to solve the problem that image recognition technology cannot recognize growth points due to occlusion and that the growth points of dicot seedlings cannot be accurately identified due to their small size, this application innovatively proposes a method that combines the replacement of the geometric center point of the detection frame with direct model recognition to improve the accuracy of growth point recognition of dicot seedlings. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following is a brief introduction to the drawings required for use in the embodiments. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other relevant drawings can be obtained based on these drawings without paying any creative work.

[0017] Figure 1 This is a flow chart of the method for identifying the growth points of dicotyledonous seedlings based on image information in this application.

[0018] Figure 2 A schematic diagram of using the method of the present application to identify growth points.

[0019] Figure 3 This is a dicotyledonous plant seedling image output after using the method of this application to identify the growing points.

[0020] Figure 4 This is the structural diagram of the YOLOv8-SEBlock-detect model for this application.

[0021] Figure 5 This is the structural diagram of this application and the YOLOv8-SEBlock-seg model. DETAILED DESCRIPTION

[0022] In order to make the purpose, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments.

[0023] This specific implementation method is specifically described using tobacco seedlings as an example.

[0024] like Figure 1 As shown, the method for identifying the growth point of a dicotyledonous plant seedling based on image information includes: S1: Optimize the YOLOv8 model backbone network. That is, embed the SEBlock (Squeeze-and-Excitation) unit after the C2f module of the YOLOv8 model backbone network to obtain the YOLOv8-SEBlock-detect model and the YOLOv8-SEBlock-seg model. The structural diagrams of the YOLOv8-SEBlock-detect model and the YOLOv8-SEBlock-seg model are as follows: Figure 4 and Figure 5 As shown in the figure. The innovative SEBlock (Squeeze-and-Excitation) channel attention mechanism is introduced into the YOLOv8 backbone network. By dynamically learning the global dependencies between feature channels, it strengthens the weight distribution of key features of growing points. Specifically, the SEBlock unit is embedded after the C2f module of the Backbone network. During feature extraction, the SEBlock unit first compresses the spatial dimensions through global average pooling to obtain channel feature statistics. It then generates a channel attention vector through a two-level fully connected layer, and finally performs an adaptive weighted fusion with the original feature map. This design effectively suppresses complex background noise and significantly improves the model's sensitivity to subtle texture features of growing points.

[0025] S2: The YOLOv8-SEBlock-detect model is trained by manually annotating the dicot seedling growth point training set, so that it can identify the dicot seedling growth point; and the YOLOv8-SEBlock-seg model is trained by manually annotating the dicot seedling outline training set, so that it can identify the entire dicot seedling; S3: Load the two trained models simultaneously and call the YOLOv8-SEBlock-detect model to directly identify and mark the growth points of all dicot seedlings in the image. S4: Call the YOLOv8-SEBlock-seg model to identify all dicot seedlings in the image and store the dicot seedling external detection frame. Determine whether there is a growth point directly identified by the YOLOv8-SEBlock-detect model within the detection frame. If not, calculate the geometric center point of the detection frame and mark it as the growth point, replacing the actual dicot seedling growth point position.

[0026] The detection frame needs to be generated by writing code after calling the model. The basic principle is to input the image → Backbone + Neck to generate a multi-scale feature map. The detection head directly predicts the bounding box on each grid cell of each feature map, converts the predicted normalized coordinates (x, y, w, h) into absolute coordinates (x1, y1, x2, y2), and calls the OpenCV library to draw a rectangular box. This rectangular box is the circumscribed rectangular box of the segmentation mask and can accurately provide the position and range of individual tobacco seedlings.

[0027] The YOLOv8-SEBlock-seg model is used to identify the entire tobacco seedling and output a rectangular detection box circumscribing the tobacco seedling. Tobacco seedlings typically have two, three, or four leaves (the number of leaves increases with growth). Their center points are highly aligned with the geometric center of their circumscribed rectangle, a property known in computer vision as "center alignment." The tobacco seedling's growth point is typically located at the center of the seedling's top-down view, very close to the geometric center of the rectangular box. Figure 2 As shown, the blue dot is the growth point detected by the model, and the red dot is the geometric center point of the rectangular box. It can be seen that the two are similar in position. Drawing on the principle of "equivalent substitution method", this application creatively proposes that the geometric center point of the rectangular box for tobacco seedling detection can replace the real growth point position when computer vision cannot directly identify it, which is used to make up for the shortcoming of computer vision in directly identifying the growth point. It can be seen that this method effectively solves the problem of being unable to identify due to occlusion and the problem that the growth point cannot be identified due to the target tobacco seedling being too small. Figure 3 As shown in the figure, this method significantly reduces the missed detection rate of directly identifying growth points through the model, which can greatly reduce the risks and losses in practical applications.

[0028] Furthermore, the method further comprises: S5: Output and save the dicot seedling image after marking the growth point.

[0029] Finally, output and save the tobacco seedling hole image after marking the growth point. Figure 3As shown, the blue ones are the growth points directly identified by computer vision, and the red ones are the growth points that are not identified by computer vision but are replaced by the geometric center point of the detection box.

[0030] Furthermore, the method further comprises: S6: Based on images of dicot seedlings with marked growing points, monitor abnormal changes in the color or morphology of the growing points to assist in the detection of early lesions; or quantify the rate of change of the growing points to assist in stress resistance assessment; or assist in forest health monitoring and quantify the progress of ecological restoration; or combine with satellite imagery to track the changes in the growing points of dicot seedlings in alpine areas with altitude to study the impact of climate on the growth of dicot seedlings.

[0031] Combined with other deep learning models, it can analyze identified growth points and monitor color and morphological abnormalities (such as yellowing and atrophy) to assist in early lesion detection. Alternatively, it can quantify the rate of change in growth points to assist in stress resistance assessment.

[0032] Furthermore, the growing point identification method can be applied to trees with similar growing point morphology. Using drone technology or drone-mounted systems to scan tree crowns, it can assist in forest health monitoring, such as analyzing the regeneration of terminal buds in trees after a fire and quantifying the progress of ecological restoration. Alternatively, combined with satellite imagery, it can track the changes in the growing point of dicot seedlings in alpine regions with altitude to study the impact of climate warming.

[0033] Furthermore, the dicotyledonous plant seedlings are tobacco seedlings, cotton seedlings, tomato seedlings, grape seedlings or forest tree seedlings.

[0034] The method of this application can be applied not only to tobacco seedlings, but also to other dicotyledonous seedlings with similar characteristics, such as cotton, tomatoes, grapes and other crops that require top pruning. This method can also be used to identify growth points and guide robots to accurately top the plants, thereby improving efficiency.

[0035] Furthermore, the geometric shape of the detection frame in this application's method is not limited to a rectangle. The rectangle was chosen based on the morphological characteristics of tobacco seedlings. For the morphological characteristics of other species, the detection frame can also be other shapes. The geometric center point replaces the growth point position to compensate for the model's shortcomings in direct recognition and reduce the error rate. This approach can be widely applied to the identification of growth points in dicot seedlings. This method can be used to construct a universal dicot seedling growth point identification framework, allowing users to quickly adapt by simply uploading a small number of new crop samples.

[0036] This application uses the single-stage object detection algorithm YOLOv8, which achieves a better balance between accuracy and efficiency through end-to-end feature learning and regression prediction. Furthermore, to address the low recognition accuracy of dicot seedling growth points due to their small size, variable morphology, and susceptibility to leaf occlusion and soil background interference, the innovative SEBlock (Squeeze-and-Excitation) channel attention mechanism is introduced into the YOLOv8 backbone network. By dynamically learning the global dependencies between feature channels, it strengthens the weight distribution of key growth point features, effectively suppresses complex background noise, and significantly improves the model's sensitivity to subtle texture features of growth points.

[0037] The above are only specific embodiments of the present application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

Claims

1. A method for identifying the growth point of a dicotyledonous plant seedling based on image information, characterized in that: include: S1: Optimize the YOLOv8 model backbone network, that is, embed the SEBlock unit after the C2f module of the YOLOv8 model backbone network to obtain the YOLOv8-SEBlock-detect model and the YOLOv8-SEBlock-seg model; S2: The YOLOv8-SEBlock-detect model is trained by manually annotating the dicot seedling growth point training set, so that it can identify the dicot seedling growth point; and the YOLOv8-SEBlock-seg model is trained by manually annotating the dicot seedling outline training set, so that it can identify the entire dicot seedling; S3: Load the two trained models simultaneously and call the YOLOv8-SEBlock-detect model to directly identify and mark the growth points of all dicot seedlings in the image. S4: Call the YOLOv8-SEBlock-seg model to identify all dicot seedlings in the image and store the dicot seedling external detection frame. Determine whether there is a growth point directly identified by the YOLOv8-SEBlock-detect model within the detection frame. If not, calculate the geometric center point of the detection frame and mark it as the growth point, replacing the actual dicot seedling growth point position.

2. The method for identifying the growth point of a dicotyledonous plant seedling based on image information according to claim 1, wherein: The method further comprises: S5: Output and save the dicot seedling image after marking the growth point.

3. The method for identifying the growth point of a dicotyledonous plant seedling based on image information according to claim 2, characterized in that: The method further comprises: S6: Based on images of dicot seedlings with marked growing points, monitor abnormal changes in the color or morphology of the growing points to assist in the detection of early lesions; or quantify the rate of change of the growing points to assist in stress resistance assessment; or assist in forest health monitoring and quantify the progress of ecological restoration; or combine with satellite imagery to track the changes in the growing points of dicot seedlings in alpine areas with altitude to study the impact of climate on the growth of dicot seedlings.

4. The method for identifying the growth point of a dicotyledonous plant seedling based on image information according to any one of claims 1 to 3, wherein: The dicotyledonous plant seedlings are tobacco seedlings, cotton seedlings, tomato seedlings, grape seedlings or forest tree seedlings.

5. The method for identifying the growth point of a dicotyledonous plant seedling based on image information according to any one of claims 1 to 3, characterized in that: The detection frame is a rectangular detection frame.

Images