Automatic monitoring method and system for whole growth period of large-scale tobacco field planting based on unmanned aerial vehicle remote sensing

Abstract

The invention provides a large-scale tobacco field planting whole growth period automatic monitoring method and system based on unmanned aerial vehicle remote sensing. According to the method, an advanced image segmentation model and a target detection algorithm are fused, and automatic extraction of a tobacco field area, tobacco plant target identification, growth stage classification judgment and planting normalization verification are realized. Firstly, unmanned aerial vehicle remote sensing images covering the whole growth period of tobacco plants are collected; then automatically completing field segmentation through an image segmentation model, and extracting a tobacco field area; identifying the tobacco plants by using the target detection model and judging the growth stages of the tobacco plants; secondly, further judging whether a planting behavior exists in the target area or not, and identifying the number of tobacco plants by adopting a model in a'planted 'state; and finally, judging whether the plants conform to a standardized planting standard or not by combining regional distribution characteristics. The tobacco plant detection and management system has the characteristics of high automation and intelligence, the tobacco plant detection and management efficiency can be remarkably improved, and powerful technical support is provided for refined decision making and large-scale popularization of modern tobacco planting.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of agricultural remote sensing detection and image recognition, in particular to a large-scale tobacco field planting full growth period automatic monitoring method and system based on unmanned aerial vehicle remote sensing. BACKGROUND

[0002] Tobacco is an important economic crop, and its planting quality and growth state have a direct impact on economic benefits. In the traditional agricultural management process, the planting monitoring of tobacco fields mainly relies on manual patrol and static image analysis, which not only has high work intensity, long cycle and high cost, but also is difficult to obtain the growth information of tobacco plants in a timely and accurate manner, especially in the rapid growth period of tobacco, which is more prone to monitoring lag.

[0003] Due to environmental factors and working conditions, manual monitoring methods often cannot cover the entire growth period, and it is difficult to achieve efficient and continuous monitoring of tobacco plants in actual production, which restricts the fine management and quality control of tobacco planting.

[0004] In recent years, with the development of unmanned aerial vehicle remote sensing and artificial intelligence technology, agricultural remote sensing monitoring based on image recognition has gradually attracted attention, but there are still many bottlenecks in the existing technology when processing large-scale tobacco field remote sensing images. On the one hand, tobacco fields are often distributed in complex background areas such as residential areas, roads, mountains and other crop plots, and existing models are easily disturbed by the background during field boundary extraction and plant identification, and have poor recognition robustness. On the other hand, many tobacco field monitoring methods can only analyze specific growth stages and cannot cover the complete growth cycle of tobacco plants from ridging, mulching, germination, growth to maturity, and cannot provide continuous monitoring data support throughout the process. In addition, most current systems still rely on manual intervention for labeling and judgment, lack of automated and standardized technical processes, and are difficult to meet the practical needs of high timeliness, high precision and low cost. SUMMARY

[0005] The present application provides a large-scale tobacco field planting full growth period automatic monitoring method and system based on unmanned aerial vehicle remote sensing, which is suitable for unmanned aerial vehicle remote sensing image automatic monitoring of large-scale tobacco fields throughout the growth period, and has key capabilities such as field automatic segmentation, tobacco plant target identification, growth stage classification and judgment, and planting specification verification. This scheme combines image segmentation models and target detection algorithms to meet the needs of remote sensing image processing in complex backgrounds, significantly improving the accuracy of tobacco plant detection. At the same time, by constructing a tobacco plant image training set covering different growth periods, continuous monitoring of tobacco plants from sowing to maturity is realized. The overall system has high automation, intelligence and practicality, and can effectively support the fine management and large-scale application of modern tobacco planting.

[0006] The first aspect of the application provides a large-scale tobacco field planting full growth period automatic monitoring method based on unmanned aerial vehicle remote sensing, comprising the following steps:

[0007] Data collection: collect remote sensing image data covering the whole tobacco planting period;

[0008] Image preprocessing: preprocessing operation is performed on the remote sensing image data;

[0009] Tobacco field information extraction: the SAM model is used to segment the field block of the preprocessed remote sensing image, and the tobacco field information is extracted;

[0010] Tobacco plant classification and identification: based on the target detection model, the extracted tobacco field information is analyzed to realize the classification and identification of the tobacco growth stage;

[0011] Planting state determination and number identification: according to the tobacco growth stage, the planting state determination is performed, and the number identification is performed if the planting state is true;

[0012] Output monitoring result: according to the field segmentation, the tobacco growth stage and the number identification, the monitoring data is exported according to the planting standard.

[0013] The second aspect of the application provides a large-scale tobacco field planting full growth period automatic monitoring system based on unmanned aerial vehicle remote sensing, comprising:

[0014] Data collection module: used for collecting remote sensing image data covering the whole tobacco planting period;

[0015] Image preprocessing module: used for preprocessing operation on the remote sensing image data;

[0016] Tobacco field information extraction module: used for segmenting the field block of the preprocessed remote sensing image by using the SAM model, and extracting the tobacco field information;

[0017] Tobacco plant classification and identification module: used for analyzing the extracted tobacco field information based on the target detection model to realize the classification and identification of the tobacco growth stage;

[0018] Planting state determination and number identification module: used for determining the planting state of the tobacco growth stage, and identifying the number if the planting state is true;

[0019] Output monitoring result module: used for structuring and arranging the field segmentation, growth stage identification and number identification, and exporting the monitoring data according to the planting standard.

[0020] In a third aspect, the present application provides a computer readable storage medium, having stored thereon a computer program, wherein the computer program, when executed by a processor, enables the method for tobacco field block identification, tobacco plant identification and planting standard analysis based on remote sensing images to be implemented.

[0021] The present application has the following advantages: the tobacco field remote sensing image intelligent identification and planting monitoring system provided by the present application can realize automatic field block extraction, intelligent identification of tobacco plant growth stages and plant number statistics in large-scale tobacco field areas; combined with actual planting density analysis, it can determine whether the planting behavior is standard, greatly improving the efficiency and accuracy of tobacco planting supervision, reducing the cost of manual field inspection and data arrangement, and providing a solid support for digital and intelligent management of agriculture. BRIEF DESCRIPTION OF DRAWINGS

[0022] Figure 1 A large-scale tobacco field planting full growth period automatic monitoring flowchart is provided for the embodiments of the present application.

[0023] Figure 2 A plant labeling example diagram is provided for the embodiments of the present application.

[0024] Figure 3 A stage identification classification diagram is provided for the embodiments of the present application.

[0025] Figure 4 A plant number identification result diagram is provided for the embodiments of the present application. DETAILED DESCRIPTION

[0026] In order to make the purpose, technical scheme and advantages of the present application clearer, the technical scheme of the present application will be described clearly and completely below in combination with the drawings in the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0027] The present application provides a large-scale tobacco field planting full growth period automatic monitoring method and system based on unmanned aerial vehicle remote sensing, a remote sensing automatic monitoring technology for tobacco planting management, suitable for full growth period monitoring of large-scale tobacco fields, especially suitable for complex terrain conditions with residential areas, roads, farmland and the like.

[0028] As shown in Figure 1 The present application is a large-scale tobacco field planting full growth period automatic monitoring method based on unmanned aerial vehicle remote sensing, comprising the following steps:

[0029] S1: Collect data, obtain UAV remote sensing images of mixed backgrounds including residential areas, farmland, roads, etc., obtain data sets of plant growth conditions, monitor the whole growth period of tobacco plants, construct tobacco field data sets at different growth periods and label them.

[0030] S2: Obtain a large-scale remote sensing tobacco field area map, use the SAM segmentation model to automatically segment the field blocks in the remote sensing image, and extract the tobacco field area.

[0031] S3: Use the data in step S1 to train the YOLO model; use the YOLO model to classify and identify the growth period of tobacco plants in the field block data extracted in step S2 (including: ridging, emergence, growth, maturity, etc.).

[0032] S4: Selectively identify the number of plants according to the identification results of step S3 to determine whether it meets the standardized planting requirements. If it is "planted", it enters the plant number identification stage. Based on the YOLOv12 model, further identify the number of tobacco plants in the current field block, and judge whether it meets the standardized planting standard combined with the regional distribution pattern.

[0033] S5: Output the planting monitoring results for external system data docking.

[0034] In a preferred embodiment, the data collection process in step S1 is as follows:

[0035] A high-definition visible light camera is mounted on a UAV to regularly take pictures of the tobacco field area in a fixed route, collecting remote sensing image data covering the entire tobacco planting period. To ensure the integrity of the time sequence, the flight time interval is set to 7 days or less, ensuring that each stage from seeding ridging, film emergence, growth period to maturity period has image records. The collected images are stored in GeoTIFF format, with geographic coordinate information. All images are classified and stored according to the collection time and field block number, forming a data structure with time and area labels. The file naming rule is: <field block number>_<plant growth period>_<date>.tif; for example: 001_mature period_20250514.tif.

[0036] Further, according to the agronomic characteristics of tobacco plants and the actual field management, the growth is divided into four stages: ridging period, emergence period, growth period, and maturity period. At the same time, the image labeling tool (labelimg) is used to manually frame the tobacco individuals in the image, as shown in Figure 2 The boundary box of the tobacco plant and its corresponding growth stage class label are labeled. The labeling result is output as a.txt file containing class, coordinate and other key information. The entire labeling process is reviewed by personnel with agronomic knowledge to ensure the accuracy of the labels.

[0037] In a preferred embodiment, the step S2 of extracting the field information specifically includes the following steps:

[0038] The core goal of this step is to accurately extract the tobacco field area from complex backgrounds such as residential areas, roads, and farmland, and to achieve automated field boundary extraction and structured expression of geographic information. First, we need to obtain the large-scale remote sensing tobacco field area image that needs to be detected through unmanned aerial photography. However, the SAM model does not support direct processing of GeoTIFF and other remote sensing image formats, so we need to preprocess the remote sensing image first. The steps are as follows:

[0039] 1. First, use the OpenCV (cv2) library to read the GeoTIFF format remote sensing image, extract and save the spatial reference information of the image (such as image resolution, projection coordinate system, upper left corner geographic coordinates, etc.), which will be used for subsequent coordinate restoration.

[0040] 2. Convert the remote sensing image format, i.e., convert the.tif format image to.png format supported by the SAM model.

[0041] Secondly, input the preprocessed image into the Segment Anything Model (SAM) deployed locally, and use its advanced image segmentation capabilities to generate masks for the field in the image, obtaining the pixel-level segmentation results of the field area. Since the SAM model outputs scattered mask regions, further boundary contour extraction and structure optimization are needed. First, use the OpenCV function to obtain the edge point set corresponding to each mask, and then use the Douglas-Peucker algorithm to simplify the preliminary boundary. This algorithm can significantly compress the number of redundant points while preserving the geometric structure characteristics, significantly improving the boundary fitting quality and subsequent data processing efficiency.

[0042] Further, restore the simplified boundary pixel coordinates to real geographic coordinates to obtain the key n points of the polygon: , ,..., , and obtain the field area by the following formula Convert the area to mu (conversion formula: 1 mu = 666.67 m²). The area information will be an important reference for subsequent tobacco plant number recognition and planting density rationality analysis.

[0043] In a preferred embodiment, the step S3 of identifying the growth stage of tobacco plants based on the YOLOv11 model specifically includes the following steps:

[0044] First, the remote sensing image dataset covering the whole growth period of tobacco plants constructed in step S1 is used as the training data for the YOLO model. The dataset has been accurately labeled according to different growth periods, as shown in Table 1, including typical stages such as ridging period, emergence period, growth period, and mature period, and reviewed by multiple artificial experts to ensure the accuracy of data labeling and stage division. Figure 3 YOLO (You Only Look Once) series models are selected for tobacco plant recognition and classification tasks in this embodiment due to their fast detection speed and high accuracy.

[0045] Further, in the model training stage, data augmentation strategies are introduced to improve the robustness and generalization ability of the model, including random rotation, color disturbance, scale scaling, and occlusion simulation, so that the model can adapt to image changes under different light, angle, and vegetation conditions. After training is completed, the YOLO model is deployed to the inference environment to process the field region images extracted by the SAM model in step S2. Since step S2 has output accurate field boundaries, this step can run the YOLO model independently within each field region to identify the target region of tobacco plants in the field block by block, and determine the growth stage based on image texture, color, and other features.

[0046] In the embodiments of the present application, the output results of the model include the position bounding box, confidence score, and growth period category of each tobacco plant. To improve the classification accuracy, a time logic verification mechanism is further introduced, which verifies and corrects the current recognition results according to the evolution trend of the stage of the field block in the historical remote sensing time sequence, thereby effectively avoiding misjudgment caused by occlusion, image blur, or abnormal posture. Finally, the spatial distribution and growth state of tobacco plants in each field block are output in a structured form, providing a reliable basis for subsequent planting density calculation and agricultural management recommendations.

[0047] In a preferred embodiment, the specific process of step S4 is as follows:

[0048] Based on the growth stage recognition results obtained in step S3, if the system determines that the current field block is in the "planted" state, i.e., the tobacco plants are in the emergence period, growth period, or mature period, etc., the plant number recognition link is automatically entered.

[0049] In the embodiments of the present application, this step aims to accurately count the number of tobacco plants in each tobacco field and conduct compliance analysis on the current planting density and plant spacing based on standardized planting standards. For this purpose, YOLOv12 model is preferably introduced, which further optimizes the detection accuracy and small target recognition ability based on YOLOv11, especially suitable for scenarios with dense arrangement of tobacco plants in the field block and significant scale variation.

[0050] Specifically, before the plant number recognition, the application will first judge whether the image size exceeds the appropriate range of model inference (such as the width and height exceeding 2048 pixels). If the image size is too large, the system will use a sliding window strategy to split the large image. By default, the image is split into a 2048x2048 pixel sub-image. This size ensures the integrity of the local image details while also considering the operational efficiency of the target detection model on small-scale images. For images with appropriate sizes, no splitting operation is performed to avoid the problem of missing edge information caused by unnecessary image cutting.

[0051] Further, after completing the image segmentation and performing the quantity recognition of the YOLOv12 model, to further improve the prediction accuracy and avoid the situation that the same plant is counted multiple times at the image segmentation boundary, the system introduces a segmentation boundary deduplication mechanism. This mechanism compares the spatial positions of the prediction results in the overlapping area, and according to the IOU (Intersection over Union) and confidence index of the prediction box, the repeatedly identified targets are merged or removed, and only the most reliable set of recognition results is retained, thereby ensuring the accuracy and uniqueness of the total plant number statistics.

[0052] In the embodiment of the present application, after obtaining the final plant number as shown in Figure 4 , the system will compare and analyze the reasonable planting density range (such as the number of tobacco plants per mu is ) specified in the empirical model, calculate the actual planting density . If does not belong to , it is determined that the current field planting behavior deviates from the standard, and the system will output a "planting density abnormal" prompt and mark the field for subsequent manual verification and intervention management.

[0053] In a preferred embodiment, the specific process of step S5 is as follows:

[0054] In the embodiment of the present application, the system will perform multi-source aggregation and structured arrangement of the recognition and analysis results obtained in the key processing steps such as field segmentation, growth stage identification, and tobacco plant number prediction, and output complete planting monitoring data files. The output result is in units of fields, and the main contents include: geographic location information (center point of longitude and latitude or projection coordinates), boundary polygon coordinates, image acquisition, identification time, growth stage classification results, number of tobacco plants in the field, unit area planting density (such as plants per mu), and whether the planting density conforms to the standardized standard and other key planting indicators.

[0055] Wherein, the judgment process of "whether it meets the standardized planting standard" is based on the boundary coordinates obtained by remote sensing image and SAM model segmentation in step S2, the real area A of each field is calculated combined with the spatial mapping relationship, and the actual unit area tobacco plant density is calculated by the total number of tobacco plants N output in step S4 . The system compares the density with the reasonable range set by the experience value of agricultural experts (such as 2600-3000 plants per mu), automatically judges whether there is abnormal planting behavior (such as over-dense, over-dilute or missing seedlings) in the current field, and gives a prompt and risk label in the result.

[0056] Finally, the system supports exporting structured monitoring results in various standardized data formats (including CSV, GeoJSON, Shapefile, etc.), and can also realize real-time data docking and sharing with agricultural supervision platform, intelligent agricultural scheduling system or tobacco planting management system through API interface, so as to comprehensively improve the digitalization, automation and intelligence level of tobacco field management, and provide scientific basis and real-time decision support for subsequent policy making and fine field operation.

[0057] Based on the same concept of the above method embodiment, the embodiment of the present application also proposes an unmanned aerial vehicle remote sensing-based automatic monitoring system for large-scale tobacco field planting in the whole growth period, comprising:

[0058] A data collection module is configured to obtain large-scale remote sensing image data taken by an unmanned aerial vehicle, and collect remote sensing image data covering the whole tobacco planting period;

[0059] An image preprocessing module is configured to perform preprocessing operation on the remote sensing image data;

[0060] A tobacco field information extraction module is configured to perform field segmentation on the preprocessed remote sensing image by using a SAM model, and extract tobacco field information;

[0061] A tobacco plant classification and identification module is configured to analyze the extracted tobacco field information based on a target detection model, and realize classification and identification of the growth stage of tobacco plants;

[0062] A planting state judgment and quantity identification module is configured to perform planting state judgment on the growth stage of tobacco plants, and perform plant number identification if the planting state is true; the plant number prediction task judges the rationality of the planting density through the prediction result;

[0063] An output monitoring result module is configured to structure and arrange the field segmentation, growth stage identification and plant number identification, export monitoring data according to the planting standard, and interface with external systems.

[0064] Based on the same concept of the method embodiments, the embodiments of the present application also propose a computer readable storage medium storing a computer program, the computer readable storage medium stores instructions, when the instructions are executed by the processor, for implementing the automatic monitoring method of large-scale tobacco field planting full growth period based on unmanned aerial vehicle remote sensing described in the present application.

[0065] In the exemplary embodiments of the present application, any implementation in the foregoing method embodiments is applicable to the embodiments of the computer readable storage medium, and will not be repeated here.

[0066] The above only describes the preferred embodiments of the present application, and it should be pointed out that for those skilled in the art, without departing from the principles of the present application, a number of improvements and refinements can be made, which should also be considered as the protection scope of the present application.

Claims

1. A method for automated monitoring of the entire growth period of large-scale tobacco field cultivation based on UAV remote sensing, characterized in that, Includes the following steps: Data collection: Collect remote sensing image data covering the entire tobacco planting cycle; Image preprocessing: Performing preprocessing operations on the remote sensing image data; Tobacco field information extraction: The SAM model is used to segment the preprocessed remote sensing images into fields and extract tobacco field information; Tobacco plant classification and identification: Based on the target detection model, the extracted tobacco field information is analyzed to achieve classification and identification of tobacco plant growth stages; Planting status determination and quantity identification: The planting status is determined based on the growth stage of the tobacco plants, and the number of plants is identified if the planting status is true. Output monitoring results: The data is structured and organized according to the field division, tobacco plant growth stage and plant number identification, and the monitoring data is exported according to the planting standards.

2. The method for automated monitoring of the entire growth period of large-scale tobacco field planting based on UAV remote sensing according to claim 1, characterized in that, The preprocessing operations include: Read remote sensing images and extract and save spatial information; The remote sensing image is then converted to a new format.

3. The method for automated monitoring of the entire growth period of large-scale tobacco field planting based on UAV remote sensing according to claim 2, characterized in that, The boundary contour extraction and structure optimization of the mask region output by the SAM model are performed, including obtaining the set of edge points corresponding to each mask and simplifying the boundary.

4. The method for automated monitoring of the entire growth period of large-scale tobacco field planting based on UAV remote sensing according to claim 3, characterized in that, The pixel coordinates of the boundary are restored to the real geographic coordinates, and the key points and field areas are obtained.

5. The method for automated monitoring of the entire growth period of large-scale tobacco field planting based on UAV remote sensing according to claim 1, characterized in that, The growth stages of the tobacco plants include the ridging stage, seedling stage, growth stage, and maturity stage.

6. The method for automated monitoring of the entire growth period of large-scale tobacco field planting based on UAV remote sensing according to claim 1, characterized in that, After identifying the number of plants, a deduplication mechanism for segmentation boundaries is introduced.

7. The method for automated monitoring of the entire growth period of large-scale tobacco field planting based on UAV remote sensing according to claim 6, characterized in that, The deduplication mechanism for segmentation boundaries merges or removes overlapping prediction boxes based on IOU and confidence level.

8. The method for automated monitoring of the entire growth period of large-scale tobacco field planting based on UAV remote sensing according to claim 1, characterized in that, The planting standards are used to predict planting abnormalities by comparing them with expert density standards.

9. A large-scale automated monitoring system for tobacco field cultivation throughout its entire growth period based on UAV remote sensing, characterized in that: include: Data collection module: used to collect remote sensing image data covering the entire tobacco planting cycle; Image preprocessing module: used to perform preprocessing operations on the remote sensing image data; Tobacco Field Information Extraction Module: Used to segment preprocessed remote sensing images into fields and extract tobacco field information using the SAM model; Tobacco plant classification and identification module: used to analyze the extracted tobacco field information based on the target detection model to realize the classification and identification of tobacco plant growth stages; Planting status determination and quantity identification module: used to determine the planting status of the tobacco plants at their growth stage, and to identify the number of plants if the planting status is true. Output monitoring results module: used to structure and organize the field segmentation, growth stage identification and plant number identification, and export the monitoring data according to the planting standards.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the computer, it implements the automated monitoring method for the entire growth period of large-scale tobacco field planting based on UAV remote sensing as described in any one of claims 1-8.

Images