Weeping forsythia flower character image recognition method based on deep learning
Through deep learning methods and the use of improved YOLO11 network structure and modules, the flower traits of Forsythia suspensa are identified, which solves the problem of poor understanding of the genetic variation of Forsythia suspensa populations and achieves efficient and accurate flower trait detection.
Patent Information
- Application Number
- CN202510884055.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-06-30
- Publication Date
- 2025-09-26
AI Technical Summary
Existing technologies have failed to effectively understand and utilize the genetic variation of wild Forsythia populations, limiting the utilization of germplasm resources and variety improvement, and lack efficient methods for identifying flower traits.
A deep learning method was used to acquire and annotate the initial dataset, perform data enhancement, establish an improved YOLO11 network structure, and use the target detection model of the C2f_SimAM attention module and the C3k2_CMUNeXt module to identify the flower traits of Forsythia suspensa.
The accuracy and efficiency of forsythia flower trait recognition are improved, support the detection of multiple traits, adapt to different occlusion conditions, enhance the focus on important features and reduce the impact of irrelevant features.
Smart Images
Figure CN120708067A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of plant trait recognition, and in particular to a method for recognizing forsythia flower trait images based on deep learning. Background Art
[0002] Forsythia suspensa, a member of the genus Forsythia in the family Oleaceae, is a commonly used Chinese medicinal herb. Its dried fruit, bitter and slightly cold in flavor, is known for its heat-clearing and detoxifying properties, reducing swelling and dispersing stagnation, and dispersing wind-heat. It is known as a "holy medicine for sores" and is commonly used in modern clinical medicine to treat acute respiratory infections, suppurative skin infections, acute nephritis, hepatitis, and meningitis. Forsythia suspensa is a temperate deciduous shrub widely distributed in the wild in my country, primarily in Shanxi, Shaanxi, Henan, Hubei, and Shandong. It is widely used in landscaping and blooms in early spring. As a common garden plant and an important Chinese medicinal herb, Forsythia suspensa possesses both high ornamental and economic value. Current research on Forsythia suspensa primarily focuses on morphology and analysis of the pharmacological effects of its active chemical components. Genetic variation within wild populations is still underdeveloped, hindering the utilization of germplasm resources and cultivar improvement. Studies have shown that Forsythia suspensa from different origins exhibits variations in its primary chemical components and antimicrobial activity, but the genetic mechanisms underlying these differences remain unclear. Understanding the genetic variation of wild populations will help us better understand the genetic mechanisms that lead to such differences and provide more references for the utilization of Forsythia medicinal resources from different origins.
[0003] Forsythia suspensa is a shallow-rooted tree species that often grows wild on sunny slopes at altitudes of 1,000 to 1,700 meters, in thickets and sparse woodlands, on partially shady slopes, in sparse woodlands in shallow mountain valleys, on riverbanks, or along forest edges. It thrives on steep cliffs or in rock crevices with poor site conditions and infertile soil. Forsythia suspensa is light-loving, cold-tolerant, drought-tolerant, and tolerant of infertility. It prefers moist soil and is sensitive to waterlogging. It is native to Gansu, Shaanxi, and Ningxia provinces in North, Southwest, and Northwest my country. It is widely distributed in the low mountainous areas of the southern Xiaolongshan Forest Region, often growing in patches or scattered patterns. With the booming landscaping and nursery stock industry and increasing demand in the industrial and medicinal markets, the area of artificial Forsythia suspensa cultivation is expanding. Summary of the Invention
[0004] The purpose of the present invention is to provide a method for recognizing the characteristics of Forsythia suspensa flowers based on deep learning to solve the problems existing in the above-mentioned background technology.
[0005] To achieve the above object, the present invention provides the following solutions: Obtain and annotate the initial dataset; Performing data enhancement on the initial data set to generate an expanded data set; Build an object detection model; Using the expanded data set to train the target detection model to generate a detection model; A detection result is generated according to the detection model and the image to be tested.
[0006] Optionally, obtaining the initial data set includes: Acquire images containing Forsythia suspensa flower traits; Annotating the images containing the characteristics of Forsythia suspensa flowers to generate an initial data set; Optionally, the trait types include petal length, petal width, petal shape, petal color, petal tip shape, pedicel length, and style type.
[0007] Optionally, the initial data set is enhanced to generate an expanded data set, specifically by performing flipping, translation, rotation and other processing on the initial data set to expand the initial data set to increase the richness of the data set and prevent the model from overfitting.
[0008] Optionally, establishing a target detection model includes: Establish the initial network structure; The initial network structure is improved to generate a target detection model.
[0009] Optionally, the initial network structure is improved to generate a target detection model, specifically by adding a C2f_SimAM attention module and a convolution module C3k2_CMUNeXt to the initial network structure to generate a target detection model.
[0010] Optionally, the C2f_SimAM attention module is implemented by adding the SimAM attention mechanism to the feature concatenation portion of the C2f module, thereby constructing the C2f_SimAM attention module and replacing the C2PSA module in the backbone network. This addition of the SimAM attention mechanism to the C2f module preserves the C2f module's ability to fuse information from different layers. Combined with the SimAM attention mechanism's ability to perceive channel and spatial information, it enhances the model's focus on different image feature dimensions and dynamically adjusts the weight of each pixel, thereby weighting important features and downweighting irrelevant ones, thereby improving the accuracy of forsythia flower recognition.
[0011] Optionally, the implementation process of the C3k2_CMUNeXt module is as follows: the CMUNeXt module with large-kernel depthwise separable convolution is applied to the parallel convolution layer in the C3k2 module to construct the C3k2_CMUNeXt module. C3k2_CMUNeXt retains the characteristics of the C3k2 module in efficiently processing data and effectively increasing the network width, thereby improving the computational efficiency of processing Forsythia flower images. At the same time, the two point-by-point convolutions that incorporate CMUNeXt's large-kernel depthwise convolution and reverse bottleneck design can improve the model's perception of Forsythia flower characteristics under different occlusion conditions. In addition, GELU activation and post-activation BatchNorm layers are used after each convolution. The definition of CMUNeXt Block is
[0012] Formula (1) represents the output feature map of the first layer in the ConvMixer block, Formula (2) represents the output feature map of the first layer after depthwise convolution, Formula (3) represents the output feature map of the first layer after pointwise convolution, represents the Batch Normalization layer, represents the GELU activation function, PointwiseConv represents the pointwise convolution layer, and DepthwiseConv represents the depthwise convolution layer.
[0013] The beneficial effects of the present invention are as follows: the present invention has practical application value for the identification of Forsythia suspensa flower traits, and the use of the improved YOLO11 network structure can detect a variety of Forsythia suspensa flower traits while maintaining a certain processing speed and accuracy. BRIEF DESCRIPTION OF THE DRAWINGS
[0014] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.
[0015] Figure 1 This is a flow chart of the forsythia flower trait recognition based on deep learning in the present invention; Figure 2 This is the backbone network structure diagram of the SCL-YOLO11 model in the present invention; Figure 3 Schematic diagram of the CMUNeXt structure in the present invention; Figure 4 It is a character diagram of the Forsythia suspensa flowers in the present invention. DETAILED DESCRIPTION
[0016] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.
[0017] Figure 1 The present invention provides a flow chart for the recognition of Forsythia flower traits based on deep learning, which mainly includes the following steps: collecting and annotating Forsythia flower trait images; performing data enhancement and expanding the data set; improving the YOLO11 model; training the improved model; identifying the Forsythia flower trait images to be detected; and obtaining the recognition results. The collected Forsythia flower trait images include petal length, petal width, petal shape, petal color, petal tip shape, pedicel length, and style type, which are subsequently annotated using labelme; data enhancement and data set expansion specifically involve flipping, translating, rotating, and other methods on the initial data set to expand the initial data set to increase the richness of the data set and prevent the model from overfitting; the initial network structure is a YOLO11 model, and improvements are made on the basis of this model, specifically by adding a SimAM attention mechanism that fuses channel attention and spatial attention in the C2f module, and applying CMUNeXt with large-kernel depthwise separable convolution in the C3k2 module. Figure 2 This is the backbone network structure diagram of the improved SCL-YOLO11 model of the present invention. Figure 3 This is a schematic diagram of the CMUNeXt structure of the present invention. After confirming the target detection model, the model needs to be trained to obtain the optimal training parameters. After the model training is completed, the forsythia flower trait image to be detected is recognized and the recognition result is obtained. Figure 4 This is a trait diagram of the Forsythia suspensa flowers of the present invention, which includes trait information of corolla tube height, pistil length, pedicel length, stamen length, petal length, petal width, and petal tip shape.
[0018] The present invention improves the initial network structure of YOLO11 in two parts, and the improved parts are described in detail below: Figure 3The implementation process of the C3k2_CMUNeXt module is as follows: the CMUNeXt module with large-kernel depthwise separable convolution is applied to the parallel convolution layer in the C3k2 module to construct the C3k2_CMUNeXt module. C3k2_CMUNeXt retains the characteristics of the C3k2 module in efficiently processing data and effectively increasing the network width, which improves the computational efficiency of processing forsythia flower images. At the same time, the integration of CMUNeXt's large-kernel depthwise convolution and two point-by-point convolutions with reverse bottleneck design can improve the model's ability to perceive the characteristics of forsythia flowers under different occlusion conditions. In addition, GELU activation and post-activation BatchNorm layers are used after each convolution. The definition of CMUNeXt Block is
[0019] Formula (1) represents the output feature map of the first layer in the ConvMixer block, Formula (2) represents the output feature map of the first layer after depthwise convolution, Formula (3) represents the output feature map of the first layer after pointwise convolution, represents the Batch Normalization layer, represents the GELU activation function, PointwiseConv represents the pointwise convolution layer, and DepthwiseConv represents the depthwise convolution layer.
[0020] The implementation process of the C2f_SimAM attention module is as follows: The SimAM attention mechanism is added to the feature concatenation portion of the C2f module to construct the C2f_SimAM attention module, replacing the C2PSA module in the backbone network. This addition of the SimAM attention mechanism to the C2f module preserves the C2f module's ability to fuse information from different layers. Combined with the SimAM attention mechanism's ability to perceive channel and spatial information, it enhances the model's focus on different image feature dimensions and dynamically adjusts the weight of each pixel, thereby overweighting important features and underweighting irrelevant ones, thereby improving the accuracy of forsythia flower recognition.
[0021] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.
[0022] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.
[0023] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The above examples are only intended to help understand the method and core concept of the present invention. At the same time, those skilled in the art will find that the specific implementation methods and application scopes may vary based on the concept of the present invention. In summary, the contents of this specification should not be construed as limiting the present invention.
Claims
1. A method for recognizing the characteristics of Forsythia suspensa flower images based on deep learning, characterized in that: include: Get the initial dataset; Performing data enhancement on the initial data set to generate an expanded data set; Build an object detection model; Using the expanded data set to train the target detection model to generate a detection model; A detection result is generated according to the detection model and the image to be tested.
2. The method for recognizing the flower traits of Forsythia suspensa based on deep learning according to claim 1, characterized in that: The obtaining of the initial data set comprises: Collect images of Forsythia suspensa flower traits and obtain information on petals, pedicels, and styles; The images of Forsythia suspensa flowers containing the three traits are annotated to generate an initial data set.
3. The method for recognizing the flower traits of Forsythia suspensa based on deep learning according to claim 2, characterized in that: The trait types include petal length, petal width, petal shape, petal color, petal tip shape, pedicel length, and style type.
4. The method for recognizing forsythia flower traits based on deep learning according to claim 1, characterized in that: The data enhancement is performed on the initial data set to generate an expanded data set. Specifically, the initial data set is processed by flipping, translating, rotating, etc. to expand the initial data set to increase the richness of the data set and prevent the model from overfitting.
5. The method for recognizing forsythia flower traits based on deep learning according to claim 1, characterized in that: The target detection model is established as follows: Establish the initial network structure; The initial network structure is improved to generate a target detection model.
6. The method for recognizing the characteristics of Forsythia suspensa flowers based on deep learning according to claim 5, characterized in that: The initial network structure is the YOLO11 network structure.
7. The method for recognizing forsythia flower traits based on deep learning according to claim 5, characterized in that: The initial network structure is improved to generate a target detection model. Specifically, the initial network structure follows the YOLO architecture, and a SimAM attention mechanism that integrates channel attention and spatial attention is added to the C2f module to replace the original C2PSA attention module, thereby improving the model's ability to focus on and weight different feature dimensions of the image. Secondly, CMUNeXt with large-kernel deep separable convolution is applied in the C3k2 module to enhance the model's perception and computational capabilities of the diverse set features of Forsythia suspensa flowers.
8. The method for recognizing forsythia flower traits based on deep learning according to claim 7, characterized in that: The implementation process for introducing the C2f_SimAM attention module involves adding the SimAM attention mechanism to the feature concatenation portion of the C2f module, thereby constructing the C2f_SimAM attention module and replacing the C2PSA module in the backbone network. This addition of the SimAM attention mechanism to the C2f module preserves the C2f module's ability to fuse information from different layers. Combined with the SimAM attention mechanism's ability to perceive channel and spatial information, it enhances the model's focus on different image feature dimensions and dynamically adjusts the weight of each pixel, thereby weighting important features and de-weighting irrelevant ones, thereby improving the accuracy of forsythia flower recognition.
9. The method for recognizing forsythia flower traits based on deep learning according to claim 7, characterized in that: The C3k2_CMUNeXt module is implemented by applying the CMUNeXt module with large-kernel depthwise separable convolutions to the parallel convolutional layers of the C3k2 module. This module, C3k2_CMUNeXt, retains the C3k2 module's efficient data processing and increased network width, improving computational efficiency in processing Forsythia flower images. Furthermore, by incorporating CMUNeXt's large-kernel depthwise convolutions and two pointwise convolutions with an inverse bottleneck design, the model's ability to perceive Forsythia flower characteristics under varying occlusion conditions is enhanced.