Coastal saline soil sunflower intelligent identification method

Abstract

The present application provides a kind of coastal saline soil sunflower intelligent identification method, it is related to agricultural intelligent monitoring technical field, comprising: obtaining the image to be measured of target area;The image to be measured is input into the sunflower intelligent identification model that is built, and the identification result is obtained;The construction method of the sunflower intelligent identification model includes: obtaining the unmanned aerial vehicle collection image of sunflower area planted in coastal saline soil field;The unmanned aerial vehicle collection image is filtered, labeled and data enhanced, and the enhanced data set is obtained;According to the enhanced data set, determine the training set and the validation set, and input the training set and the validation set into the improved YOLOv7tiny network identification model for training and verification, and the sunflower intelligent identification model that is built is obtained.The present application can not only be applied in coastal saline soil, but also can be popularized to tea chrysanthemum and other chrysanthemum plants precise identification, to optimize the field management and yield estimation of sunflower in coastal saline soil provides effective data support.

Description

Technical Field

[0001] This invention relates to the field of agricultural intelligent monitoring technology, and in particular to an intelligent identification method for sunflowers in coastal saline-alkali land. Background Technology

[0002] Sunflowers are annual herbaceous plants widely used for food, medicine, and ornamental purposes. They are drought-resistant, salt-tolerant, tolerant of poor soil, and highly adaptable, making them a top choice for ecological restoration of coastal saline-alkali lands. Sunflowers play a crucial role in the cultivation of coastal saline-alkali lands in my country. The flowering period of sunflowers is a critical stage in their growth cycle; accurately identifying the number of flower heads and buds is essential for yield estimation, screening of high-quality germplasm resources, determination of phenological periods, and optimization of field management. However, manual statistical methods are time-consuming, labor-intensive, and prone to human error.

[0003] Traditional digital image processing techniques for crop identification face various limitations in field environments, such as varying lighting conditions, shading, and soil background interference. These factors reduce accuracy and make it difficult to meet practical needs. However, in recent years, the rapid development of deep learning and UAV remote sensing technologies has provided effective technical means for the precise management and counting of crops in large-scale farmland. Target detection can be divided into two main categories: two-stage detectors (such as R-CNN), which require two feature extractions, resulting in a large network structure, high computational resource consumption, and slow processing speed; and single-stage detectors (such as YOLO and SSD), which only require one feature extraction, are more efficient and are therefore widely used for real-time detection.

[0004] While current single-stage detection algorithms improve detection efficiency to some extent, they may also suffer from some accuracy loss. Therefore, improving the accuracy of sunflower identification in coastal saline-alkali lands without sacrificing efficiency is a pressing practical problem. The complex field environment of coastal saline-alkali lands, coupled with the interference of lighting and motion blur in drone photography, and the dense planting and mutual shading of sunflowers, present significant challenges to identification. Current target detection algorithms cannot adapt well to changes in receptive field size in different locations when dealing with geometric deformations, resulting in poor feature extraction capabilities for different sunflower states, such as flowering, semi-flowering, and budding stages. Furthermore, there is currently no intelligent identification method specifically designed for sunflowers grown in coastal saline-alkali lands, either domestically or internationally. This intelligent identification method can not only be applied to coastal saline-alkali lands but can also be extended to the accurate identification of other Asteraceae plants such as tea chrysanthemum, providing effective data support for optimizing field management and yield estimation of sunflowers in coastal saline-alkali lands. Summary of the Invention

[0005] In order to overcome the shortcomings of the existing technology, the purpose of this invention is to provide a method for intelligent identification of sunflowers in coastal saline-alkali land.

[0006] To achieve the above objectives, the present invention provides the following solution:

[0007] A method for intelligent identification of sunflowers in coastal saline-alkali land includes:

[0008] Acquire the image of the target region;

[0009] The image to be tested is input into the constructed Sunflower intelligent recognition model to obtain the recognition result; the construction method of the Sunflower intelligent recognition model includes:

[0010] To acquire drone-collected images of sunflower fields planted in coastal saline-alkali land;

[0011] The images acquired by the UAV are filtered, labeled, and augmented to obtain an augmented dataset;

[0012] The training set and validation set are determined based on the augmented dataset, and the training set and validation set are input into the improved YOLOv7tiny network recognition model for training and validation, so as to obtain the constructed Sunflower intelligent recognition model.

[0013] Preferably, the construction process of the improved YOLOv7tiny network recognition model includes:

[0014] Obtain the original YOLOv7tiny model;

[0015] In the ELAN structure of the backbone network of the YOLOv7tiny model, the original CBL ordinary convolutional module with a kernel size of 3 is replaced with deformable convolution DCNv2. The outputs of different levels are fused by the Concat module and then introduced into the SimAM attention mechanism to obtain the improved DCNv2-SimAM-ELAN structure.

[0016] The improved DCNv2-SimAM-ELAN structure replaces the last three ELAN structures in the backbone network of the YOLOv7tiny model, and a SimAM attention mechanism is added after the first ELAN structure.

[0017] The loss function is replaced by the WIoU loss function to obtain the improved YOLOv7tiny network recognition model.

[0018] Preferably, the images acquired by the UAV are filtered, labeled, and augmented to obtain an augmented dataset, including:

[0019] The images collected by the drone were filtered, including visible light images of drones from different varieties, different growth stages, different lighting conditions, and different planting densities.

[0020] Use software image annotation tools to annotate the selected images;

[0021] Use image processing tools to perform random data augmentation on the labeled images.

[0022] Preferably, the ratio of the training set to the validation set is 8:2.

[0023] Preferably, the data annotation results include the sunflower's flowering state, semi-flowering state, and bud state.

[0024] Preferably, the data enhancement method includes randomly changing the image brightness and contrast.

[0025] Preferably, the software image annotation tool is LabelImg.

[0026] Preferably, the software image data enhancement tool is OpenCV.

[0027] According to specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0028] This invention provides a method for intelligent recognition of sunflowers in coastal saline-alkali land, comprising: acquiring a test image of a target area; inputting the test image into a constructed intelligent sunflower recognition model to obtain a recognition result; the method for constructing the intelligent sunflower recognition model comprises: acquiring drone-collected images of sunflower fields planted in coastal saline-alkali land; filtering, labeling, and data augmenting the drone-collected images to obtain an augmented dataset; determining a training set and a validation set based on the augmented dataset, and inputting the training set and the validation set into an improved YOLOv7tiny network recognition model for training and validation to obtain the constructed intelligent sunflower recognition model. This invention can not only be applied to coastal saline-alkali land, but can also be extended to the accurate recognition of other Asteraceae plants such as tea chrysanthemum, providing effective data support for optimizing field management and yield estimation of sunflowers in coastal saline-alkali land. Attached Figure Description

[0029] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0030] Figure 1A flowchart of the method provided in an embodiment of the present invention;

[0031] Figure 2 This is a schematic diagram of the technical route provided in the embodiments of the present invention;

[0032] Figure 3 This is a schematic diagram of the YOLOv7tiny ELAN structure provided in an embodiment of the present invention;

[0033] Figure 4 A schematic diagram of deformable convolution DCNv2 provided in an embodiment of the present invention;

[0034] Figure 5 This is a schematic diagram of the SimAM attention mechanism provided in an embodiment of the present invention;

[0035] Figure 6 This is an example diagram of the DCNv2-SimAM-ELAN structure provided in an embodiment of the present invention;

[0036] Figure 7 A schematic diagram of the improved YOLOv7tiny network structure provided in an embodiment of the present invention. Detailed Implementation

[0037] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0038] The purpose of this invention is to provide a smart identification method for sunflowers in coastal saline-alkali land. This method can not only be applied to coastal saline-alkali land, but can also be extended to the accurate identification of other Asteraceae plants such as tea chrysanthemum, providing effective data support for optimizing field management and yield estimation of sunflowers in coastal saline-alkali land.

[0039] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0040] Figure 1 The method flowchart provided in the embodiments of the present invention is as follows: Figure 1 As shown, this invention provides a method for intelligent identification of sunflowers in coastal saline-alkali land, comprising:

[0041] Step 100: Obtain the image of the target region to be tested;

[0042] Step 200: Input the image to be tested into the constructed Sunflower Intelligent Recognition Model to obtain the recognition result; the construction method of the Sunflower Intelligent Recognition Model includes:

[0043] Step 201: Acquire drone images of sunflower fields planted in coastal saline-alkali land;

[0044] Step 202: Filter, label, and augment the images acquired by the UAV to obtain an augmented dataset;

[0045] Step 203: Determine the training set and validation set based on the augmented dataset, and input the training set and validation set into the improved YOLOv7tiny network recognition model for training and validation to obtain the constructed Sunflower intelligent recognition model.

[0046] like Figure 2 As shown, the technical approach of this embodiment is as follows:

[0047] S1. Using a drone equipped with a visible light camera to collect images of sunflower fields in coastal saline-alkali land. In this embodiment, sunflowers grown in coastal saline-alkali land were used as the research object. From July to August 2023, data was collected from two sunflower planting plots within the National Crop Variety Demonstration and Evaluation Base of the Yellow River Delta Agricultural High-tech Zone, National Saline-Alkali Land Comprehensive Utilization Technology Innovation Center, Guangrao County, Dongying City, Shandong Province. A total of 24 different sunflower varieties were planted in the two plots. Before the drone flight, the flight path between the two plots was planned, with a flight altitude of 10 meters, a 60% overlap in the flight direction, and a 60% overlap in the lateral direction. During the shooting, the visible light camera was set to automatic mode, and the shooting time was concentrated between 10:00 AM and 4:00 PM.

[0048] S2. In this example, the acquired images are cropped to 640*640 pixels, and high-quality images are selected for data annotation. The annotations include three labels for the sunflower images: flowering state, half-flowering state, and bud state, which are "sunflower", "half-open", and "bud" respectively. The coordinates (x, y) of the annotation center point, as well as the width w and height h of the annotation box, are also annotated.

[0049] S3. In this embodiment, OpenCV image processing tools are used for random data augmentation. The data augmentation methods include: cropping, translation, changing brightness, adding noise, rotating angle, mirroring and random combinations of cutout, in order to enhance the robustness of the model. The augmented dataset is divided into training set and validation set in an 8:2 ratio.

[0050] S4. The augmented training set was input into the original training set to train the original YOLOv7tiny model. The following parameter settings were used: Stochastic Gradient Descent (SGD) as the optimizer, 200 training epochs, 32 batch size, 0.01 initial learning rate, 0.937 momentum, and 0.0005 weight decay. The validation set was loaded for testing.

[0051] S5. In this embodiment, the structure of the improved YOLOv7tiny network mainly consists of the backbone network portion of the YOLOv7tiny network, which has four ELAN structures. Based on the YOLOv7tiny model, this invention designs a new ELAN structure, named DCNv2-SimAM-ELAN. The original YOLOv7tiny ELAN structure is as follows: Figure 3 As shown, Figure 6 The DCNv2-SimAM-ELAN structure (DS-ELAN) designed for this invention.

[0052] Specifically, in the original ELAN structure of the YOLOv7tiny backbone network, the original CBL ordinary convolutional module (Conv3×3) with a kernel size of 3 is replaced with a deformable convolutional network DCNv2 (Deformable Convolution Network v2), thereby more effectively capturing changes in the morphology of sunflower inflorescences in UAV remote sensing images. After fusing features from different levels through the Concat module, the SimAM attention mechanism is introduced to enhance feature extraction and integration capabilities. For example... Figure 6 As shown, the improved DCNv2-SimAM-ELAN structure replaces the last three ELAN structures in the original YOLOv7tiny backbone network, outputting ELAN structures with feature maps of 180×80, 40×40, and 20×20 respectively. A SimAM attention mechanism is added after the first ELAN structure (outputting a 160×160 feature map). Simultaneously, the model's loss function is replaced from CIoU (Complete-Intersection over Union) to WIoU (Wise-Intersection over Union), allowing the model to focus on high-quality bounding boxes, improving model accuracy and accelerating convergence.

[0053] In the DCNv2-SimAM-ELAN architecture, deformable convolution operations are performed on a two-dimensional plane, as shown in the structure... Figure 4 As shown, compared to traditional fixed sampling grids, deformable convolutional networks employ learnable sampling grids and introduce position offset and shape transformation parameters. These parameters predict the offset of sampling points, allowing the position and shape of the convolutional kernel to be dynamically adjusted. This adaptively changes the sampling position and focuses on the semantic feature points and geometric key points of the target. This approach better adapts to irregular target shapes and positions, enabling the model to more effectively capture more fine-grained spatial features. Deformable convolution is represented as:

[0054]

[0055] Where Δp n These are learnable parameters, floating-point data, obtained from the image through ordinary convolution; they represent the positional offset of the convolution kernel. DCNv2 is an improved version of deformable convolutional networks, reducing computational complexity and irrelevant information. The determination method for DCNv2 is as follows:

[0056]

[0057] DCNv2 adds weights to each sampling point, where Δm k These are the weighting coefficients.

[0058] In the DCNv2-SimAM-ELAN architecture, SimAM is a parameter-free attention mechanism, and its overall structure is as follows: Figure 5 As shown, unlike traditional channel attention mechanisms, it generates attention weights with full 3D weights. Full 3D weights allow for more fine-tuning of features. In visual tasks, to better achieve attention, neurons with significant spatial inhibition effects should be assigned higher priority. This is achieved by defining an energy function that measures the linear separability between neurons. This helps in identifying important neurons, and the energy function has an analytical solution, thus accelerating weight computation. The input feature map has C channels and a size of H×W, and the final output feature map is obtained by multiplying the weights of the sought neurons with the features in the original feature map. The minimum energy of a neuron can be expressed as:

[0059]

[0060] in x i ∈X represents a neuron cell. This formula shows that the lower the energy, the greater the difference between neuron t and surrounding neurons, and the higher its importance. Finally, the sigmoid function is used to suppress excessively high values ​​to satisfy the definition of the attention mechanism, as shown in the formula:

[0061]

[0062] S6. The improved YOLOv7tiny network recognition model (e.g.) Figure 7 The model is trained (as shown), and a validation set is loaded for testing to obtain the model's prediction results.

[0063] S7. Compare and analyze the model experimental results from S4 and S6, and analyze the detection results. The accuracy of the model predictions is measured using precision (P, %), recall (R, %), and mean average precision (mAP, %), calculated using the following formulas:

[0064]

[0065]

[0066]

[0067]

[0068] Among them, X TP This indicates the number of sunflowers correctly detected as being in the flowering, semi-flowering, or budding stage; X FP This indicates the number of sunflowers incorrectly detected as being in the flowering, semi-flowering, or budding stage; X FN denoted as the number of sunflowers in bloom, half-bloom, and bud stages that were missed in the image; mAP is the average value of P for all R values ​​between 0 and 1; C is the number of detected categories. The target of this method includes sunflowers in bloom, half-bloom, and bud stages, so C is 3.

[0069] Table 1. Recognition accuracy of the original YOLOv7tiny model on the validation set.

[0070]

[0071] Table 2 shows the recognition accuracy of the improved YOLOv7tiny model on the validation set.

[0072]

[0073] As shown in Tables 1 and 2, the original YOLOv7tiny model performed poorly in detecting the semi-open state subclass. The improved model improved precision by 6.5% to 90%, recall by 1.3% to 87.3%, and AP by 3.8% to 92.4% for the semi-open state subclass. AP was improved by 0.6% and 0.3% for the bud and flowering states, respectively. The overall model precision reached 91.1%, recall reached 89%, and mAP reached [missing value]. @0.5 It reached 92.5%, representing improvements of 2.6%, 1.1%, and 1.6% respectively compared to the original YOLOv7tiny model.

[0074] S8. Using image post-processing, based on the model's recognition results, visualize the sunflower's flowering, semi-flowering, and budding states in the form of rectangular boxes on the image to highlight the recognition results.

[0075] Furthermore, this embodiment combines DCNv2 and SimAM attention mechanisms into the ELAN structure of the YOLO backbone network, designing a DCNv2-SimAM-ELAN module specifically to address the challenges in recognizing sunflowers grown in coastal saline-alkali land. This module dynamically adjusts the position of the convolutional kernels to better adapt to sunflower targets of different scales and shapes. It also introduces a SimAM attention mechanism with full 3D weights, enabling the model to better enhance important features and suppress unimportant features when processing sunflower inflorescence detection tasks. These improvements enhance the model's perceptual capabilities and feature extraction quality. Simultaneously, the original YOLOv7tiny loss function (Loss) is replaced with the WIoU loss function, allowing the model to focus more efficiently on high-quality bounding boxes. This helps improve the accuracy of object detection, thereby reducing false positives and false negatives, further enhancing the accuracy of sunflower recognition.

[0076] In summary, the key technical improvements of this invention lie in enhancing the model's perception capabilities, feature extraction quality, and accurate target detection methods. These improvements are significant for sunflower identification in coastal saline-alkali lands, and will help provide more accurate data, offering more reliable support for field management and yield estimation of sunflowers in coastal saline-alkali lands.

[0077] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0078] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A coastal saline soil sunflower intelligent identification method, characterized in that, The method comprises the following steps: acquiring a to-be-tested image of a target area; inputting the to-be-tested image into a constructed sunflower intelligent recognition model to obtain a recognition result; the construction method of the sunflower intelligent recognition model comprises the following steps: acquiring unmanned aerial vehicle (UAV) collected images of a sunflower area planted in a coastal saline-alkali field; performing screening, labeling and data enhancement on the UAV collected images to obtain an enhanced data set; determining a training set and a verification set according to the enhanced data set, and inputting the training set and the verification set into an improved YOLOv7tiny network recognition model for training and verification to obtain the constructed sunflower intelligent recognition model; the construction process of the improved YOLOv7tiny network recognition model comprises the following steps: acquiring an original YOLOv7tiny model; in an ELAN structure in a backbone network in the YOLOv7tiny model, replacing a CBL ordinary convolution module with a convolution kernel size of 3 with a deformable convolution DCNv2, introducing a SimAM attention mechanism after feature fusion of outputs of different levels through a Concat module to obtain an improved DCNv2-SimAM-ELAN structure; replacing the improved DCNv2-SimAM-ELAN structure with the last three ELAN structures in the backbone network in the YOLOv7tiny model, and adding a SimAM attention mechanism after the first ELAN structure; replacing a loss function from CIoU to WIoU to obtain the improved YOLOv7tiny network recognition model. 2.The coastal saline soil sunflower intelligent identification method according to claim 1, characterized in that, performing screening, labeling and data enhancement on the UAV collected images to obtain an enhanced data set, comprising: performing screening on the UAV collected images, including UAV visible light images of different varieties, different growth stages, different light conditions and different planting densities; using a software image labeling tool to label the screened images; using an image processing tool to perform random data enhancement on the labeled images. 3.The coastal saline soil sunflower intelligent identification method according to claim 1, characterized in that, The ratio of the training set to the verification set is 8:

2.

4. The coastal saline soil sunflower intelligent identification method according to claim 2, characterized in that, The labeling result comprises a flowering state, a semi-flowering state and a bud state of the sunflower.

5. The coastal saline soil sunflower intelligent identification method according to claim 2, characterized in that, The data enhancement method comprises randomly changing image brightness and contrast. 6.The coastal saline soil sunflower intelligent identification method according to claim 2, characterized in that, The software image labeling tool is LabelImg.

7. The coastal saline soil sunflower intelligent identification method according to claim 2, characterized in that, The image processing tool is OpenCV.

Images