A farmland protection inspection system and method based on Beidou short message communication

The farmland protection patrol system based on BeiDou short message communication utilizes multi-rotor drones and point-based YOLOX-Nano models for image recognition and data transmission, solving the problem of low monitoring efficiency in farmland protection in remote mountainous areas and achieving high-frequency, precise farmland protection.

CN118785368BActive Publication Date: 2024-11-08CIVIL AVIATION FLIGHT UNIV OF CHINA
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411267518.1
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-09-11
Publication Date
2024-11-08
Estimated Expiration
2044-09-11

AI Technical Summary

Technical Problem

Existing technologies suffer from low monitoring efficiency, high costs, and insufficient signal coverage in farmland protection in remote mountainous areas, making it difficult to effectively meet the needs of farmland protection.

Method used

A farmland protection patrol system based on BeiDou short message communication is adopted. It utilizes a multi-rotor UAV equipped with an optical camera, a BeiDou short message communication module, a Raspberry Pi and a microcontroller. Combined with a point-based YOLOX-Nano model, it performs image recognition and processing to achieve data transmission and location information integration. The system obtains location information through a BeiDou positioning module to realize signal and information transmission between the UAV and the ground monitoring station.

Benefits of technology

To improve patrol efficiency under complex climatic conditions, reduce waste of manpower and financial resources, achieve continuous and high-frequency monitoring of arable land, ensure that monitoring is not affected by remote sensing monitoring and unstable cellular network signals, and provide high-precision arable land protection solutions.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118785368B_ABST
    Figure CN118785368B_ABST
Patent Text Reader

Abstract

The invention discloses a farmland protection inspection system and method based on Beidou short message communication, which relates to the technical field of farmland protection. The system includes a multi-rotor drone, which is equipped with an optical camera, a Beidou short message communication module, a raspberry pie, a single-chip microcomputer and a Beidou positioning module. The raspberry pie is connected to the optical camera and the single-chip microcomputer respectively, and the single-chip microcomputer is connected to the Beidou short message communication module and the Beidou positioning module respectively; the optical camera is used to obtain real-time image data of farmland; the Beidou short message communication module is used to realize data transmission; the point based yoloX‑Nano model in the raspberry pie is used to perform image recognition and processing on real-time image data of farmland; the single-chip microcomputer is used to integrate location information and non-agricultural information; and the Beidou positioning module is used to obtain location information. The invention solves the problem that the existing methods have limitations to varying degrees and are difficult to effectively meet the needs of farmland protection.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of farmland protection, and in particular to a farmland protection inspection system and method based on Beidou short message communication. Background Art

[0002] With the acceleration of urbanization and the development of industry, farmland resources are facing serious challenges. For example, the occupation, pollution and over-cultivation of farmland are common, resulting in a continuous reduction in the area of ​​farmland and a decline in its quality. Especially in remote mountainous areas, due to the vast land and sparse population, the signal coverage of communication base stations is relatively insufficient, and the complex terrain and changeable climate make farmland protection more difficult. Traditional methods of farmland protection can no longer meet the needs.

[0003] At present, farmland protection mainly relies on the following methods: satellite remote sensing monitoring, "field manager system" grid management and fixed-point monitoring in concentrated and contiguous areas. Although satellite remote sensing monitoring can provide a wide range of farmland information, it is not effective in cloudy and foggy areas, and the monitoring cycle is relatively long; the "field manager system" grid management relies on a large amount of manpower, is inefficient and costly, and is not ideal in fragmented farmland with poor communication signals; the installation and maintenance costs of concentrated and contiguous fixed-point monitoring equipment are high in mountainous areas, and the coverage is limited. These methods have limitations to varying degrees in practical applications and are difficult to effectively meet the needs of farmland protection. Summary of the invention

[0004] In view of the above-mentioned deficiencies in the prior art, the present invention provides a farmland protection inspection system and method based on Beidou short message communication, which solves the problem that the existing methods have limitations to varying degrees and are difficult to effectively meet the needs of farmland protection.

[0005] In order to achieve the above-mentioned invention object, the technical solution adopted by the present invention is: a farmland protection inspection system based on Beidou short message communication, including a multi-rotor drone, the multi-rotor drone is equipped with an optical camera, a Beidou short message communication module, a Raspberry Pi, a single-chip microcomputer and a Beidou positioning module, the Raspberry Pi stores a point based yoloX-Nano model, the Raspberry Pi is respectively connected to the optical camera and the single-chip microcomputer, and the single-chip microcomputer is respectively connected to the Beidou short message communication module and the Beidou positioning module;

[0006] The optical camera is used to obtain real-time image data of cultivated land;

[0007] The Beidou short message communication module is used to realize data transmission;

[0008] The point based yoloX-Nano model in the Raspberry Pi is used to perform image recognition and processing on real-time image data of cultivated land;

[0009] The single chip microcomputer is used to integrate the location information and non-agricultural information;

[0010] The Beidou positioning module is used to obtain location information.

[0011] The beneficial effects of the above scheme are as follows: the present invention provides a method for patrolling farmland protection based on Beidou short message communication, and analyzes images through a lightweight and efficient point based yoloX-Nano model. YoloX-Nano adopts an anchor-free design and combines Mosaic data enhancement to maintain high detection accuracy in an environment with limited computing resources. The Beidou short message communication module is used to realize the signal and information transmission between the drone and the ground monitoring station, which can effectively improve the patrol efficiency under complex climatic conditions such as remote mountainous areas with weak digital signal cellular network signal reception capabilities, eliminate the waste of manpower and financial resources caused by problems such as remote sensing monitoring being affected or cellular network signals being poor, and solve the problems that the existing methods have limitations to varying degrees and are difficult to effectively meet the needs of farmland protection.

[0012] In addition, the present invention also adopts a technical solution: a method for patrolling farmland protection based on Beidou short message communication, comprising the following steps:

[0013] S1: Use a multi-rotor drone to patrol the farmland according to the cruise route and obtain real-time image data of the farmland based on an optical camera;

[0014] S2: Use the point based yoloX-Nano model to perform image recognition and processing on real-time image data of cultivated land to obtain non-agricultural information;

[0015] S3: Use the Beidou positioning module to obtain the current location information, and use the single-chip microcomputer to integrate the location information and non-agricultural information;

[0016] S4: Use the Beidou short message communication module to transmit the integrated information to the terminal monitoring center, and use the terminal monitoring center to analyze and process the information, formulate corresponding protection plans, and complete farmland protection inspections based on Beidou short message communication.

[0017] The beneficial effect of the above scheme is: the present invention provides a farmland protection inspection method based on Beidou short message communication, analyzes the image through the point based yoloX-Nano model, and realizes the signal and information transmission between the UAV and the ground monitoring station through the Beidou short message communication module. It can effectively improve the inspection efficiency under complex climatic conditions such as mountainous areas with weak signal reception capabilities, and realize continuous and high-frequency monitoring of farmland to ensure that the monitoring is not affected by remote sensing technology interference or unstable cellular network signals, which solves the problems that the existing methods have limitations to varying degrees and are difficult to effectively meet the needs of farmland protection.

[0018] Furthermore, S2 includes the following sub-steps:

[0019] S2-1: Collect the dataset of cultivated land images and annotate the dataset using annotation software;

[0020] S2-2: Divide and preprocess the labeled data set, and input the preprocessed data set into the pointbased yoloX-Nano model for training;

[0021] S2-3: The image data acquired by the optical camera in real time is transmitted to the trained point based yoloX-Nano model for recognition and processing to obtain non-agricultural information.

[0022] The beneficial effect of the above further scheme is that by training the model, the model parameters are continuously optimized so that it can accurately identify non-agricultural areas.

[0023] Furthermore, the point based yoloX-Nano model is supervised by point-level annotation information in weakly supervised learning, and the yolo detection head of the point based yoloX-Nano model is decoupled to perform different tasks in stages, including completing the regression task in the first stage and assisting in the classification task in the second stage with the help of a cross-stage convolution module.

[0024] Furthermore, the loss function of the point based yoloX-Nano model includes classification loss function, target loss function and positioning loss function.

[0025] The beneficial effect of the above further scheme is that the present invention adopts classification loss function, target loss function and positioning loss function for the point based yoloX-Nano model to further improve the model accuracy.

[0026] Furthermore, the classification loss function for:

[0027]

[0028] in, is the actual label, is the probability value output by the model.

[0029] Furthermore, the objective loss function for:

[0030]

[0031] in, The model predicts that the sample belongs to The probability of the class, is the total number of categories, is the balance factor, is the adjustment factor.

[0032] Furthermore, the positioning loss function for:

[0033]

[0034] in, is the intersection-over-union ratio of the predicted box and the true box, is the number of images, is the number of point labels, is the number of proposal boxes, is the NOM score, is the Euclidean distance, is the center point of the prediction box, is the center point of the real frame, is the diagonal distance of the minimum enclosed area that contains both the predicted box and the true box, is the balance factor, is the normalized distance score, is the weight coefficient, is the shape consistency measure, For the Image No. Point labels The normalized score of the proposal box;

[0035]

[0036]

[0037] in, and are the width and height of the real frame respectively, and are the width and height of the prediction box respectively;

[0038] The NOM score The calculation formula is:

[0039] .

[0040] The beneficial effect of the above further solution is: in order to balance the impact of pseudo-frame labels, the present invention adopts an improved positioning loss function. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 The figure is a flow chart of a farmland protection inspection method based on Beidou short message communication.

[0042] Figure 2 Flowchart for point based yoloX-Nano model recognition and processing of captured images. DETAILED DESCRIPTION

[0043] The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments.

[0044] Embodiment 1, a farmland protection inspection system based on Beidou short message communication, including a multi-rotor drone, the multi-rotor drone is equipped with an optical camera, a Beidou short message communication module, a Raspberry Pi, a single-chip microcomputer and a Beidou positioning module, the Raspberry Pi stores a point based yoloX-Nano model, the Raspberry Pi is respectively connected to the optical camera and the single-chip microcomputer, and the single-chip microcomputer is respectively connected to the Beidou short message communication module and the Beidou positioning module;

[0045] The optical camera is used to obtain real-time image data of cultivated land;

[0046] The Beidou short message communication module is used to realize data transmission;

[0047] The point based yoloX-Nano model in the Raspberry Pi is used to perform image recognition and processing on real-time image data of cultivated land;

[0048] The single chip microcomputer is used to integrate the location information and non-agricultural information;

[0049] The Beidou positioning module is used to obtain location information.

[0050] In one embodiment of the present invention, the hardware construction of the drone includes: using Raspberry Pi and ESP32 microcontroller for programming to realize the route planning, positioning and message sending of the drone; installing a camera connected to the Raspberry Pi on the drone to shoot the ground situation and transmit it to the Raspberry Pi for processing; installing a Beidou positioning module Pro (the model of the Beidou positioning module Pro in this embodiment is BDQH-MIX-CA) to obtain the real-time longitude and latitude information of the drone to ensure accuracy and prevent the loss of the drone; using the Beidou-3 communication module to upload the identified ground conditions and longitude and latitude information to the satellite; using a lithium battery as a power supply device to ensure normal operation under various weather conditions.

[0051] Embodiment 2, as Figure 1 As shown, a farmland protection inspection method based on Beidou short message communication includes the following steps:

[0052] S1: Use a multi-rotor drone to patrol the farmland according to the cruise route and obtain real-time image data of the farmland based on an optical camera;

[0053] S2: Use the point based yoloX-Nano model to perform image recognition and processing on real-time image data of cultivated land to obtain non-agricultural information;

[0054] In this embodiment, non-agriculturalization refers to the use of cultivated land for production and operation activities other than agricultural production. Non-agriculturalization information refers to the illegal occupation of cultivated land for non-agricultural project construction, such as factory construction, house construction, road construction, garbage dumping, etc.

[0055] S3: Use the Beidou positioning module to obtain the current location information, and use the single-chip microcomputer to integrate the location information and non-agricultural information;

[0056] S4: Use the Beidou short message communication module to transmit the integrated information to the terminal monitoring center, and use the terminal monitoring center to analyze and process the information, formulate corresponding protection plans, and complete farmland protection inspections based on Beidou short message communication.

[0057] In one embodiment of the present invention, a multi-rotor drone equipped with an optical camera and a Beidou short message transceiver module is used to obtain image data of cultivated land in real time and perform preliminary processing; in the Beidou short message communication module, data transmission between the drone and the monitoring center is realized through short message communication, ensuring stable data transmission even in a signal-free area, and recording the flight trajectory of the drone; in the image recognition and processing process, the improved point based yoloX-Nano model is used to process the cultivated land images collected in real time to identify non-agricultural crop information in the cultivated land. The model extracts image features through a convolutional neural network and performs target detection and classification; when the monitoring center analyzes and processes the data, the cultivated land is comprehensively monitored and evaluated in combination with a geographic information system (GIS), and a corresponding protection plan is formulated; the flight trajectory data recorded by the Beidou short message system can be used to evaluate and optimize the inspection scope and effect, ensuring the comprehensiveness and accuracy of the inspection process.

[0058] S2 includes the following sub-steps:

[0059] S2-1: Collect the dataset of cultivated land images and annotate the dataset using annotation software;

[0060] In this embodiment, a multi-rotor drone equipped with an optical camera was used to photograph the farmland, and two hundred images of different crops and non-crops were collected. In order to improve the generalization ability of the model, the data set includes farmland images taken in different seasons, at different times, and under different lighting conditions;

[0061] Use labelimg annotation software to manually annotate the crops and non-crops in the cultivated land in the dataset, with the label names Crop and Non-crop respectively;

[0062] S2-2: Divide and preprocess the labeled data set, and input the preprocessed data set into the pointbased yoloX-Nano model for training;

[0063] In this embodiment, the pictures and annotation files are divided into a training set, a validation set, and a test set, and corresponding label files are generated respectively. Before training the model, the data set is preprocessed, including adjusting the image size and performing data enhancement. The prepared data set is input into the point based yoloX-Nano model for training. During the training process, the model parameters are optimized so that it can accurately identify non-agricultural areas. In order to avoid overfitting of the model, different training strategies and rounds are used for training. The specific steps are as follows: 50 rounds, 100 rounds, and 150 rounds of training rounds are set respectively, and comparative experiments are carried out. Finally, the model with the highest verification accuracy is selected as the final prediction model;

[0064] S2-3: The image data acquired by the optical camera in real time is transmitted to the trained point based yoloX-Nano model for recognition and processing to obtain non-agricultural information.

[0065] The recognition process of this model is as follows Figure 2 As shown in the figure, the model identifies the non-agricultural areas in the image, marks them with boxes, and outputs results containing six data items: xmin, ymin, xmax, ymax, confidence, and class. The output results mark the locations of non-agricultural areas with boxes, and organize information such as coordinates and confidence into a data set.

[0066] The point based yoloX-Nano model is supervised by point-level annotation information in weakly supervised learning, and the yolo detection head of the point based yoloX-Nano model is decoupled to perform different tasks in stages, including completing the regression task in the first stage and assisting the classification task in the second stage with the help of a cross-stage convolution module.

[0067] The loss function of the point based yoloX-Nano model includes a classification loss function, a target loss function and a positioning loss function.

[0068] Classification loss function for:

[0069]

[0070] in, is the actual label (0 or 1, the real event did not happen or happened), It is the probability value output by the model (the probability of predicting the positive class).

[0071] Objective loss function for:

[0072]

[0073] in, The model predicts that the sample belongs to The probability of the class, is the total number of categories, is the balance factor, is the adjustment factor.

[0074] Localization loss function for:

[0075]

[0076] in, is the intersection-over-union ratio of the predicted box and the true box, is the number of images, is the number of point labels, is the number of proposal boxes, is the NOM score, is the Euclidean distance, is the center point of the prediction box, is the center point of the real frame, is the diagonal distance of the minimum enclosed area that contains both the predicted box and the true box, is the balance factor, is the normalized distance score, is the weight coefficient, is the shape consistency measure, For the Image No. Point labels The normalized score of the proposal box;

[0077]

[0078]

[0079] in, and are the width and height of the real frame respectively, and are the width and height of the prediction box respectively;

[0080] The NOM score The calculation formula is:

[0081] .

[0082] In one embodiment of the present invention, the location information and non-agricultural information are integrated and uploaded as follows: the serial port is initialized, the longitude and latitude information of the UAV obtained by the Beidou-3 positioning module is read, and stored in a variable; the data set output by the improved point based yoloX-Nano model is integrated with the longitude and latitude information obtained by the Beidou positioning module, and the integrated data is sent to the single-chip microcomputer using the XCOM 2.0 serial port assistant; the single-chip microcomputer processes the received txt text, integrates the non-agricultural information with the longitude and latitude, and sends the processed data to the satellite through the communication module to realize real-time inspection and non-agricultural area positioning functions; a communication protocol including longitude and latitude, text, category information, confidence, minimum x coordinate, minimum y coordinate, maximum x coordinate, and maximum y coordinate is designed, and the integrated message is sent to the Beidou satellite according to the GB2312 character encoding to report the inspection results in real time.

[0083] The present invention utilizes an unmanned aerial vehicle platform to cover a wide area, especially in mountainous areas with complex terrain and inconvenient transportation, thereby significantly improving inspection efficiency; the present invention combines the Beidou short message system to realize real-time data transmission and flight trajectory recording, thereby ensuring stable data transmission even in areas without signals; the present invention adopts an improved point based yoloX-Nano model for image recognition and processing, thereby accurately identifying non-agricultural crop information in cultivated land, thereby improving the accuracy and reliability of monitoring; the present invention analyzes and processes data through a monitoring center, combines with a geographic information system (GIS) for comprehensive monitoring and evaluation, formulates corresponding protection plans, and ensures that cultivated land resources are effectively protected; the present invention uses the flight trajectory data recorded by the Beidou short message system to evaluate and optimize the inspection scope and effect, thereby ensuring the comprehensiveness and accuracy of the inspection process.

[0084] Those skilled in the art will appreciate that the embodiments described herein are intended to help readers understand the principles of the present invention, and should be understood that the protection scope of the present invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific variations and combinations that do not deviate from the essence of the present invention based on the technical revelations disclosed by the present invention, and these variations and combinations are still within the protection scope of the invention.

Claims

1. A method for patrolling farmland protection based on Beidou short message communication, characterized in that: The following steps are involved: S1: Use a multi-rotor drone to patrol the farmland according to the cruise route and obtain real-time image data of the farmland based on an optical camera; S2: Use the point based yoloX-Nano model to perform image recognition and processing on real-time image data of cultivated land to obtain non-agricultural information; S3: Use the Beidou positioning module to obtain the current location information, and use the single-chip microcomputer to integrate the location information and non-agricultural information; S4: Use the Beidou short message communication module to transmit the integrated information to the terminal monitoring center, and use the terminal monitoring center to analyze and process the information, formulate corresponding protection plans, and complete the farmland protection inspection based on Beidou short message communication; The S2 includes the following sub-steps: S2-1: Collect the dataset of cultivated land images and annotate the dataset using annotation software; Collect image data including different crops and non-agricultural crops, wherein the data set includes images of cultivated land taken in different seasons, different times and different lighting conditions; Use labeling software to label crops and non-crops, and the label names are Crop and Non-crop respectively; S2-2: Divide and preprocess the labeled data set, and input the preprocessed data set into the point basedyoloX-Nano model for training; The preprocessing includes adjusting the image size and performing data enhancement; S2-3: The image data acquired by the optical camera in real time is transmitted to the trained point based yoloX-Nano model for recognition and processing to obtain non-agricultural information; When the trained point based yoloX-Nano model identifies a non-agricultural area in the image data, a box is selected and marked, and the output result marks the location of the non-agricultural area with a box, and the coordinates and confidence levels are organized into a data set; The point based yoloX-Nano model is supervised by point-level annotation information in weakly supervised learning, the yolo detection head of the point based yoloX-Nano model is decoupled, and different tasks are performed in stages, including completing the regression task in the first stage and assisting the classification task in the second stage with the help of the cross-stage convolution module; The loss function of the point based yoloX-Nano model includes a classification loss function, a target loss function and a positioning loss function; The classification loss function for: in, is the actual label, is the probability value output by the model; The objective loss function for: in, The model predicts that the sample belongs to The probability of the class, is the total number of categories, is the balance factor, is the adjustment factor; The positioning loss function for: in, is the intersection-over-union ratio of the predicted box and the true box, is the number of images, is the number of point labels, is the number of proposal boxes, is the NOM score, is the Euclidean distance, is the center point of the prediction box, is the center point of the real frame, is the diagonal distance of the minimum enclosed area that contains both the predicted box and the true box, is the balance factor, is the normalized distance score, is the weight coefficient, is the shape consistency measure, For the Image No. Point labels The normalized score of the proposal box; in, and are the width and height of the real frame respectively, and are the width and height of the prediction box respectively; The NOM score The calculation formula is: 。 2. A farmland protection inspection system based on the farmland protection inspection method based on Beidou short message communication as claimed in claim 1, characterized in that: It includes a multi-rotor drone, which is equipped with an optical camera, a Beidou short message communication module, a Raspberry Pi, a single-chip microcomputer and a Beidou positioning module. The Raspberry Pi stores a point based yoloX-Nano model, the Raspberry Pi is connected to the optical camera and the single-chip microcomputer respectively, and the single-chip microcomputer is connected to the Beidou short message communication module and the Beidou positioning module respectively; The optical camera is used to obtain real-time image data of cultivated land; The Beidou short message communication module is used to realize data transmission; The point based yoloX-Nano model in the Raspberry Pi is used to perform image recognition and processing on real-time image data of cultivated land; The single chip microcomputer is used to integrate the location information and non-agricultural information; The Beidou positioning module is used to obtain location information.