Invasive plant detection system based on helmet type image acquisition equipment

Abstract

The invention discloses an invasive plant detection system based on helmet-type image acquisition equipment. The system comprises a background server, an invasive plant image recognition algorithm and the helmet-type image acquisition equipment, the helmet type image acquisition device is composed of goggles, a camera, a main control core, a GPS positioner, a sensor module, a storage battery, a helmet, a wireless communication module and an alarm arranged on the front portion of the helmet, and all the modules are connected through circuits. Compared with a traditional manual inspection mode, the method can quickly and accurately identify the involved invasive plant species, and can perform detection in a wider area. Therefore, the inspection cost and workload can be greatly reduced, the burden of researchers is reduced, and the working efficiency is improved. And meanwhile, the system has the advantages of good real-time performance, wide application range, convenience in use and the like, and is suitable for monitoring and protecting road invasive plants which are not suitable for unmanned aerial vehicles.

Description

Technical Field

[0001] The present invention relates to the field of plant protection, and in particular to an invasive plant detection system based on a helmet-type image acquisition device, which is particularly suitable for areas where drones are not applicable. Background Art

[0002] Invasive plants, such as exotic harmful plants, are one of the main causes of ecological damage and species extinction. The commonly used method for invasive plant detection is to use drones for vegetation remote sensing detection, but there are the following problems with using drones in many wild areas:

[0003] 1) In densely forested areas, drones are easily obstructed, and their flight altitude and range are also limited;

[0004] 2) Drones require professional operators, are costly and prone to failure;

[0005] 3) Plants are scarce, target plants are blocked, and their proportion in the image is too small, making plant recognition difficult.

[0006] Therefore, there is a need for an efficient and accurate invasive plant detection method that can make up for the shortcomings of drone detection. Summary of the invention

[0007] Therefore, the present invention provides an invasive plant detection system based on a helmet-type image acquisition device to solve the above-mentioned technical problems. It can detect invasive plants around roads and in forests on the ground, thereby improving the efficiency of ecological and environmental protection personnel in identifying and removing invasive plants.

[0008] In order to achieve the above-mentioned purpose, the present invention adopts the following technical scheme: an invasive plant detection system based on a helmet-type image acquisition device, the system includes three parts: a background server, an invasive plant image recognition algorithm and a helmet-type image acquisition device, which work together to realize automatic and real-time detection of invasive plants; the invasive plant image recognition algorithm is based on a target detection model to identify different types of invasive plants; the helmet-type image acquisition device is composed of goggles, a camera, a main control core, a GPS locator, a sensor module, a battery, a helmet, a wireless communication module, and an alarm arranged at the front of the helmet, and the modules are connected through circuits. The helmet is structurally reinforced and comfortably designed to provide a good wearing experience and has sufficient stability to support various attached modules; the goggles are integrated on the helmet to provide basic eye protection functions and support the display of key detection information.

[0009] Preferably, the background server is used for image processing and storage of large-scale image data, and for in-depth analysis. The server can be an ordinary computer (such as a laptop or desktop) or a cloud server. The use of a cloud server can provide better scalability and computing power, adapt to large-scale data processing needs, and also include a data analysis module and a model update module to perform in-depth analysis and long-term storage of received image data, and regularly update and optimize the invasive plant detection model based on newly collected plant image data to improve the recognition accuracy of the system. The invasive plant detection model is optimized and has the characteristics of fast response and low memory usage, and supports dynamic adjustment of detection thresholds to adapt to the recognition needs of different environments and plant species; the background server also supports data synchronization and collaborative detection of multiple helmet-type image acquisition devices, which is convenient for teamwork and large-scale ecological monitoring.

[0010] Preferably, the camera includes a high-resolution camera and a night vision camera for all-weather plant image acquisition, can adapt to a variety of lighting conditions, and achieve all-weather plant image acquisition; its photo mode and video mode both support real-time video stream transmission and static image capture, and have autofocus and anti-shake functions to ensure image clarity.

[0011] Preferably, the main control core module includes an embedded computer and an edge computing unit, which processes large-scale image data in real time, has efficient computing power and low power consumption characteristics, can process large-scale image data in real time, and support local data storage, so that key tasks can still be completed when the network is unstable. The embedded computer is an embedded development board, on which an invasive plant image recognition algorithm is deployed, which can run efficiently under low computing resources, support automatic detection and identification in offline mode, and can be connected to external storage devices through an expansion interface to enhance data processing capabilities. The main control core has efficient computing power and low power consumption characteristics, can support real-time detection of small batches of invasive plants, and complete local processing when the network is unstable.

[0012] Preferably, the sensor module includes a temperature sensor, a humidity sensor and a light intensity sensor to monitor the surrounding environmental parameters in real time, and associate these data with plant image information to improve the accuracy, reliability and robustness of detection; at the same time, the sensor module has a self-calibration function to ensure the accuracy of data collection.

[0013] Preferably, the wireless communication module includes Bluetooth, Wi-Fi and 5G communication modules, has multiple communication modes, and can automatically switch according to the network environment to ensure the stability and real-time of data transmission, and ensure the stable connection between the helmet-type device and the background server; the module also supports communication collaboration between multiple devices, and realizes data sharing and collaborative detection between multiple helmet devices.

[0014] Preferably, the GPS locator adopts a high-precision GPS chip and supports multi-star system reception to accurately obtain and record the location information of the collected plant images to provide more accurate positioning services. The recorded longitude and latitude information is uploaded to the background server together with the plant image and the detection results, supporting the visualization analysis of spatial data and the generation of invasive plant distribution maps. These location information are associated with the detection results for subsequent management and monitoring.

[0015] Preferably, the alarm can emit sound and light alarms and vibration feedback to ensure that users can obtain alarm information in a complex environment in a timely manner; at the same time, the alarm has a custom function for alarm conditions, and can automatically adjust the alarm intensity and form according to the type and severity of the invasive plants detected. When invasive plants are detected, the alarm will sound an alarm through sound, light signal or vibration to remind users to take timely action. The intensity and form of the alarm can be adjusted according to the type and severity of the detected plants.

[0016] Preferably, the battery adopts a high-energy-density replaceable lithium battery pack, which has the characteristics of high energy density and long life, and supports fast charging and multiple protection functions, ensuring the power supply of the equipment during long-term field operations and ensuring the reliable operation of the equipment in various environments.

[0017] Preferably, the invasive plant detection model is optimized, has the characteristics of fast response and low memory usage, and supports dynamic adjustment of detection thresholds to adapt to the identification requirements of different environments and plant species; the invasive plant detection model is based on a lightweight deep convolutional neural network, has efficient image feature extraction capabilities and strong robustness, and can maintain stable detection accuracy under various environmental conditions; the training process of the invasive plant detection model includes data set construction, model selection, parameter tuning and performance evaluation. A comparative analysis was conducted with mainstream target detection models such as R-CNN, and the optimal model was finally determined as the basic model for invasive plant detection in combination with the lightweight and real-time requirements in practical applications. The trained invasive plant detection model needs to be converted into a format to adapt to the operating environment of the embedded development board. The specific steps include converting the PyTorch model into the ONNX format, and then converting the ONNX format model into an engine file adapted to the embedded development board to ensure the efficient operation of the model in a low-resource environment. The invasive plant detection model is installed on the embedded development board of a helmet-type image acquisition device. The user wears the helmet-type image acquisition device, and the camera captures the surrounding plant images, and the captured images are input into the pre-trained invasive plant detection network for real-time analysis. The recognition results are prompted by an alarm, and the GPS positioning module records the current location information for further analysis and decision-making. The invasive plant detection model can efficiently and accurately identify a variety of invasive plants, and support large-scale automated inspections, thereby significantly reducing labor costs and improving the efficiency and quality of ecological environmental protection work.

[0018] Workflow:

[0019] The user wears a helmet-type image acquisition device, and the camera captures the surrounding plant images in real time. The image signal is transmitted to the main control core for preliminary processing. The processed image is input into the pre-trained invasive plant detection network, which analyzes and identifies the plant species in the image in real time. The recognition result is fed back to the user in a timely manner through the alarm. At the same time, the GPS positioning module records the location information of the plant image for subsequent management.

[0020] In addition, users can also transmit image data to the background server through the wireless communication module for further detection on the server. The server compares the image data and determines whether there are invasive plants through model analysis. If an abnormality is found, the system will issue an alarm and record the relevant location information to assist managers in subsequent ecological monitoring and decision-making.

[0021] Beneficial effects of the present invention:

[0022] The present invention utilizes a helmet-type image acquisition device to detect invasive plants, which can achieve all-round and real-time monitoring of roads and mountainous forest areas, reduce the workload of manual inspections required in traditional methods, and make up for the shortcomings of drones in detecting invasive plants. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 It is a structural schematic diagram of an invasive plant detection system based on a helmet-type image acquisition device of the present invention;

[0024] Figure 2 It is a flow chart of invasive plant detection of the present invention.

[0025] Among them: goggles-1, camera-2, main control core-3, GPS locator-4, sensor module-5, battery-6, helmet-7, wireless communication module-8. DETAILED DESCRIPTION

[0026] In order to further explain the technical solution of the present invention, it is described in detail below through specific embodiments.

[0027] like Figure 1 As shown, the present invention provides an invasive plant detection system based on a helmet-type image acquisition device, which includes three parts: a background server, an invasive plant image recognition algorithm and a helmet-type image acquisition device; the invasive plant image recognition algorithm is based on a target detection model to identify different types of invasive plants; the helmet-type image acquisition device is composed of goggles 1, a camera 2, a main control core 3, a GPS locator 4, a sensor module 5, a battery 6, a helmet 7, a wireless communication module 8, and an alarm arranged at the front of the helmet 7, and the modules are connected through a circuit. The helmet 7 is structurally strengthened and comfortably designed to provide a good wearing experience, and has sufficient stability to support various attached modules, ensuring the stable installation of each module and the long-term wearing comfort of the device; the goggles are integrated on the helmet, providing basic eye protection functions, and supporting the display of key detection information, integrating eye protection functions, protecting the user's eyes from the influence of the external environment, and displaying key detection information when necessary.

[0028] In the present invention, the backend server is used for image processing and storage of large-scale image data, and for in-depth analysis. The server can be an ordinary computer (such as a notebook or desktop) or a cloud server. The use of a cloud server can provide better scalability and computing power, adapt to large-scale data processing needs, and also include a data analysis module and a model update module, which perform in-depth analysis and long-term storage of received image data, and regularly update and optimize the invasive plant detection model based on the newly collected plant image data to improve the recognition accuracy of the system. The invasive plant detection model is optimized and has the characteristics of fast response and low memory usage, and supports dynamic adjustment of the detection threshold to adapt to the recognition needs of different environments and plant species; the backend server also supports data synchronization and collaborative detection of multiple helmet-type image acquisition devices, which is convenient for teamwork and ecological monitoring over a large range.

[0029] The camera 2 in the present invention includes a high-resolution camera and a night vision camera for all-weather plant image acquisition, which can adapt to a variety of lighting conditions and realize all-weather plant image acquisition; its photo mode and video mode both support real-time video stream transmission and static image capture, and have automatic focus and anti-shake functions to ensure image clarity. The camera 2 can capture image information of the surrounding environment in real time, and transmit the image signal to the main control core for processing through the internal circuit.

[0030] The main control core module 3 in the present invention includes an embedded computer and an edge computing unit, which processes large-scale image data in real time, has efficient computing power and low power consumption characteristics, can process large-scale image data in real time, and support local data storage, so that key tasks can still be completed when the network is unstable. The embedded computer is an embedded development board, on which an invasive plant image recognition algorithm is deployed, which can run efficiently under low computing resources, support automatic detection and recognition in offline mode, and can be connected to an external storage device through an expansion interface to enhance data processing capabilities. The main control core has efficient computing power and low power consumption characteristics, can support real-time detection of small batches of invasive plants, and complete local processing when the network is unstable, that is, work independently to ensure the continuity of detection tasks.

[0031] The sensor module 5 in the present invention includes a temperature sensor, a humidity sensor and a light intensity sensor, which monitor the surrounding environmental parameters in real time and associate these data with plant image information to improve the accuracy, reliability and robustness of detection; at the same time, the sensor module has a self-calibration function to ensure the accuracy of data collection.

[0032] The wireless communication module 8 in the present invention includes Bluetooth, Wi-Fi and 5G communication modules, has multiple communication modes, and can automatically switch according to the network environment to ensure the stability and real-time performance of data transmission, and ensure the stable connection between the helmet-type device and the background server; the module also supports communication collaboration between multiple devices, and realizes data sharing and collaborative detection between multiple helmet devices.

[0033] The GPS locator 4 in the present invention adopts a high-precision GPS chip and supports multi-star system reception to accurately obtain and record the location information of the collected plant images to provide more accurate positioning services. The recorded longitude and latitude information is uploaded to the background server together with the plant image and the detection results, supporting the visualization analysis of spatial data and the generation of invasive plant distribution maps. These location information are associated with the detection results for subsequent management and monitoring.

[0034] The alarm in the present invention can emit sound and light alarms and vibration feedback to ensure that users can obtain alarm information in a complex environment in a timely manner; at the same time, the alarm has a custom function for alarm conditions, and can automatically adjust the alarm intensity and form according to the type and severity of the invasive plants detected. When invasive plants are detected, the alarm will sound an alarm through sound, light signals or vibrations to remind users to take timely action. The intensity and form of the alarm can be adjusted according to the type and severity of the detected plants.

[0035] The storage battery 6 in the present invention adopts a high-energy-density replaceable lithium battery pack, which has the characteristics of high energy density and long life, and supports fast charging and multiple protection functions, ensuring the power supply of the equipment during long-term field operations and ensuring the reliable operation of the equipment in various environments.

[0036] The invasive plant detection model in the present invention is optimized, has the characteristics of fast response and low memory usage, and supports dynamic adjustment of detection thresholds to adapt to the identification requirements of different environments and plant species; the invasive plant detection model is based on a lightweight deep convolutional neural network, has efficient image feature extraction capabilities and strong robustness, and can maintain stable detection accuracy under various environmental conditions; the training process of the invasive plant detection model includes data set construction, model selection, parameter tuning and performance evaluation. A comparative analysis was conducted with mainstream target detection models such as R-CNN, and the optimal model was finally determined as the basic model for invasive plant detection in combination with the lightweight and real-time requirements in practical applications. The trained invasive plant detection model needs to be converted into a format to adapt to the operating environment of the embedded development board. The specific steps include converting the PyTorch model into the ONNX format, and then converting the ONNX format model into an engine file adapted to the embedded development board to ensure the efficient operation of the model in a low-resource environment. The invasive plant detection model is installed on the embedded development board of a helmet-type image acquisition device. The user wears the helmet-type image acquisition device, and the camera captures the surrounding plant images, and the captured images are input into the pre-trained invasive plant detection network for real-time analysis. The recognition results are prompted by an alarm, and the GPS positioning module records the current location information for further analysis and decision-making. The invasive plant detection model can efficiently and accurately identify a variety of invasive plants, and support large-scale automated inspections, thereby significantly reducing labor costs and improving the efficiency and quality of ecological environmental protection work.

[0037] Model training and deployment

[0038] Dataset creation: We used images of invasive plants taken by mobile phones and sports cameras, combined with web crawlers and video frame extraction, to build an image dataset containing a variety of invasive plants. We cleaned, cropped, and named the collected images, and used the Labelimg tool to generate label files to ensure the quality and diversity of the dataset.

[0039] Model training: In the Pytorch framework, transfer learning and fine-tuning training methods are used to train Yolo, SSD, Faster R-CNN and other target detection models. Through transfer learning, the model can expand its application scope and enhance its robustness when the sample data set is insufficient. Fine-tuning training speeds up the training of detection models for newly added plant categories and improves recognition accuracy.

[0040] Model optimization: By comparing the performance of various target detection models, the appropriate model is selected as the basis, and the model is optimized for the identification difficulties of specific invasive plants to improve the detection speed and accuracy of the model.

[0041] Model deployment: Use TensorRT inference engine technology to accelerate the trained model and adapt it to the embedded computing environment. The model deployment path is: convert the PyTorch model to ONNX format, and then convert it to the TensorRT engine file. The deployed model performs real-time inference through embedded devices and is tested and optimized in actual applications.

[0042] Specific workflow

[0043] like Figure 2 The figure shows the workflow of the invasive plant detection system, which includes the following steps:

[0044] Equipment wearing: Users wear helmet-type image acquisition equipment to ensure its stability and comfort so that the camera can effectively capture images of the surrounding environment;

[0045] Image acquisition: The built-in camera of the helmet captures the environment image in real time, and the image is transmitted to the main control core through the circuit for preliminary processing. The processed image can be transmitted to the background server through the wireless communication module;

[0046] Image preprocessing: The backend server preprocesses the received images, including denoising and image enhancement, to improve the accuracy of detection;

[0047] Invasive plant detection: The pre-trained detection model infers the pre-processed image, identifies the invasive plants in the image, and generates detection results;

[0048] GPS positioning and recording: The built-in GPS positioning module of the helmet device obtains the latitude and longitude information of the current image and associates it with the detection results for storage, supporting subsequent management and tracking;

[0049] Alarm prompt: After detecting invasive plants, the alarm will sound or vibrate to remind the user to take necessary measures;

[0050] Control measures: Based on the location information generated by the system, managers can locate invasive plants and take control measures, such as spraying or mechanical removal, to curb the spread of plants.

[0051] The above description is only a preferred example of the present invention and is not intended to limit the present invention. Although the present invention is described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or replace some of the technical features therein by equivalents. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. An invasive plant detection system based on a helmet-type image acquisition device, characterized in that: The system comprises three parts: a background server, an invasive plant image recognition algorithm and a helmet-type image acquisition device; the invasive plant image recognition algorithm is based on a target detection model and identifies different types of invasive plants; the helmet-type image acquisition device comprises goggles (1), a camera (2), a main control core (3), a GPS locator (4), a sensor module (5), a battery (6), a helmet (7), a wireless communication module (8), and an alarm arranged at the front of the helmet (7), and each module is connected through a circuit.

2. The invasive plant detection system based on a helmet-type image acquisition device according to claim 1, characterized in that: The background server is used for image processing and data storage, and also includes a data analysis module and a model update module, which perform in-depth analysis and long-term storage on the received image data, and regularly update and optimize the invasive plant detection model according to the newly collected plant image data.

3. The invasive plant detection system based on a helmet-type image acquisition device according to claim 1, characterized in that: The camera (2) comprises a high-resolution camera and a night-vision camera for collecting plant images around the clock.

4. The invasive plant detection system based on a helmet-type image acquisition device according to claim 2, characterized in that: The main control core module (3) comprises an embedded computer and an edge computing unit, and processes large-scale image data in real time. The embedded computer is an embedded development board, on which an invasive plant image recognition algorithm is deployed, supports automatic detection and recognition in an offline mode, and can be connected to an external storage device through an expansion interface.

5. The invasive plant detection system based on a helmet-type image acquisition device according to claim 1, characterized in that: The sensor module (5) comprises a temperature sensor, a humidity sensor and a light intensity sensor, which monitors the surrounding environment parameters in real time and associates these data with the plant image information.

6. The invasive plant detection system based on a helmet-type image acquisition device according to claim 1, characterized in that: The wireless communication module (8) includes Bluetooth, Wi-Fi and 5G communication modules.

7. The invasive plant detection system based on a helmet-type image acquisition device according to claim 1, characterized in that: The GPS locator (4) uses a high-precision GPS chip and supports multi-star system reception. The recorded longitude and latitude information is uploaded to the backend server together with the plant image and the detection result.

8. The invasive plant detection system based on a helmet-type image acquisition device according to claim 1, characterized in that: The alarm can emit sound and light alarms and vibration feedback. The alarm has the function of customizing alarm conditions and can automatically adjust the alarm intensity and form according to the type and severity of the invasive plants detected.

9. The invasive plant detection system based on a helmet-type image acquisition device according to claim 1, characterized in that: The storage battery (6) is a replaceable lithium battery pack.

10. The invasive plant detection system based on a helmet-type image acquisition device according to claim 4, characterized in that: The invasive plant detection model is optimized to support dynamic adjustment of detection thresholds to adapt to the identification requirements of different environments and plant species; the invasive plant detection model is based on a lightweight deep convolutional neural network; the training process of the invasive plant detection model includes data set construction, model selection, parameter tuning and performance evaluation, and by comparing and analyzing mainstream target detection models such as Yolo, SSD, and Faster R-CNN, combined with the lightweight and real-time requirements in practical applications, the optimal model is finally determined as the basic model for invasive plant detection; the trained invasive plant detection model needs to be format converted to adapt to the operating environment of the embedded development board; the specific steps include converting the PyTorch model into ONNX format, and then converting the ONNX format model into an engine file adapted to the embedded development board; the invasive plant detection model is installed on the embedded development board of the helmet-type image acquisition device.

Images