CN114442623B - Agricultural machinery operation track Tian Lu segmentation method based on space-time diagram neural network - Google Patents
Agricultural machinery operation track Tian Lu segmentation method based on space-time diagram neural network Download PDFInfo
- Publication number
- CN114442623B CN114442623B CN202210065003.8A CN202210065003A CN114442623B CN 114442623 B CN114442623 B CN 114442623B CN 202210065003 A CN202210065003 A CN 202210065003A CN 114442623 B CN114442623 B CN 114442623B
- Authority
- CN
- China
- Prior art keywords
- track
- space
- point
- time
- agricultural machinery
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000010586 diagram Methods 0.000 title claims abstract description 49
- 238000000034 method Methods 0.000 title claims abstract description 41
- 230000011218 segmentation Effects 0.000 title claims abstract description 34
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 19
- 230000000694 effects Effects 0.000 claims abstract description 15
- 239000011159 matrix material Substances 0.000 claims description 23
- 238000012546 transfer Methods 0.000 claims description 15
- 238000010276 construction Methods 0.000 claims description 14
- 238000004220 aggregation Methods 0.000 claims description 12
- 230000002776 aggregation Effects 0.000 claims description 12
- 238000012549 training Methods 0.000 claims description 11
- 238000013527 convolutional neural network Methods 0.000 claims description 10
- 230000001133 acceleration Effects 0.000 claims description 9
- 238000004422 calculation algorithm Methods 0.000 claims description 9
- 230000006870 function Effects 0.000 claims description 9
- 238000004364 calculation method Methods 0.000 claims description 5
- 230000009466 transformation Effects 0.000 claims description 5
- 230000004913 activation Effects 0.000 claims description 4
- 238000000605 extraction Methods 0.000 claims description 4
- 238000005065 mining Methods 0.000 claims description 4
- 230000008859 change Effects 0.000 claims description 3
- 238000005096 rolling process Methods 0.000 claims description 3
- 239000000463 material Substances 0.000 abstract description 3
- 241000209094 Oryza Species 0.000 description 9
- 235000007164 Oryza sativa Nutrition 0.000 description 9
- 241000209140 Triticum Species 0.000 description 9
- 235000021307 Triticum Nutrition 0.000 description 9
- 235000009566 rice Nutrition 0.000 description 9
- 238000004140 cleaning Methods 0.000 description 6
- 238000003306 harvesting Methods 0.000 description 6
- 238000005516 engineering process Methods 0.000 description 5
- 238000002474 experimental method Methods 0.000 description 5
- 230000008569 process Effects 0.000 description 5
- 238000003066 decision tree Methods 0.000 description 4
- 238000007637 random forest analysis Methods 0.000 description 4
- 238000013136 deep learning model Methods 0.000 description 3
- 238000011161 development Methods 0.000 description 3
- 230000018109 developmental process Effects 0.000 description 3
- 238000012952 Resampling Methods 0.000 description 2
- 238000012271 agricultural production Methods 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 2
- 238000010835 comparative analysis Methods 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 230000001419 dependent effect Effects 0.000 description 2
- 238000003709 image segmentation Methods 0.000 description 2
- 238000005070 sampling Methods 0.000 description 2
- 230000003068 static effect Effects 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 230000036962 time dependent Effects 0.000 description 2
- 238000012795 verification Methods 0.000 description 2
- 241001124569 Lycaenidae Species 0.000 description 1
- 230000002159 abnormal effect Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000009826 distribution Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 239000000575 pesticide Substances 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000000630 rising effect Effects 0.000 description 1
- 238000009331 sowing Methods 0.000 description 1
- 238000005507 spraying Methods 0.000 description 1
- 238000013106 supervised machine learning method Methods 0.000 description 1
- 230000002123 temporal effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05D—SYSTEMS FOR CONTROLLING OR REGULATING NON-ELECTRIC VARIABLES
- G05D1/00—Control of position, course, altitude or attitude of land, water, air or space vehicles, e.g. using automatic pilots
- G05D1/02—Control of position or course in two dimensions
- G05D1/021—Control of position or course in two dimensions specially adapted to land vehicles
- G05D1/0212—Control of position or course in two dimensions specially adapted to land vehicles with means for defining a desired trajectory
- G05D1/0217—Control of position or course in two dimensions specially adapted to land vehicles with means for defining a desired trajectory in accordance with energy consumption, time reduction or distance reduction criteria
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A40/00—Adaptation technologies in agriculture, forestry, livestock or agroalimentary production
- Y02A40/10—Adaptation technologies in agriculture, forestry, livestock or agroalimentary production in agriculture
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Aviation & Aerospace Engineering (AREA)
- Radar, Positioning & Navigation (AREA)
- Remote Sensing (AREA)
- Automation & Control Theory (AREA)
- Traffic Control Systems (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses an agricultural machinery operation track Tian Lu segmentation method based on a space-time diagram neural network, belonging to the technical field of computer network application. According to the method, a space-time diagram neural network based on a time-space relation is adopted to divide an effective operation track of an agricultural machine, the Beidou positioning acquisition module arranged on the agricultural machine is utilized to acquire agricultural machine track data, and the space-time diagram neural network is constructed by utilizing association information among different track points in the agricultural machine track data so as to excavate more abundant track point characteristic information, so that a better track division effect is achieved; according to the invention, the space-time diagram neural network aiming at the agricultural machinery operation track points is created by utilizing the space-time correlation information contained in the agricultural machinery track data, so that the agricultural machinery operation track is automatically segmented into farmlands and roads, and a large amount of manpower and material resources are saved.
Description
Technical Field
The invention belongs to a space-time diagram neural network-based agricultural machinery operation track Tian Lu segmentation method in the technical field of computer network application.
Background
The existing effective operation track segmentation method and technology of the agricultural machinery mainly have the following defects:
(1) The segmentation method based on farmland boundaries comprises the following steps: firstly, manually collecting related farmland boundaries; secondly, when the agricultural machine runs, whether the agricultural machine enters a farmland or not is automatically judged according to the real-time positioning information of the agricultural machine, so that the aim of dividing the operation track Tian Lu is fulfilled. However, manual collection of the boundaries of farmlands consumes manpower and material resources, cannot adapt to the development trend of less humanization and no humanization of the operation of the agricultural machinery, has the condition of false report missing report in manual collection, and is not beneficial to statistics and supervision of the effective operation track of the agricultural machinery.
(2) The segmentation method based on the remote sensing image comprises the following steps: given a remote sensing image of a running area of the agricultural machinery, and carrying out road segmentation by adopting an image segmentation method. The method is greatly limited by the remote sensing image data itself. For example, low resolution of the remote sensing image may reduce the image segmentation effect, thereby reducing the segmentation accuracy.
(3) The segmentation method based on density clustering comprises the following steps: aiming at the characteristic that the densities of track points on farmlands and roads are different, a clustering method is adopted to segment tracks Tian Lu; however, the density characteristics are too single to effectively divide the agricultural work trajectory Tian Lu. For example, the track point density at the road traffic lights is often high, resulting in areas that are misidentified as farmlands, reducing segmentation accuracy.
(4) Segmentation method based on traditional machine learning: the track points are classified ("farmland" vs. "roads") using a traditional fully supervised machine learning method. In the track point classification process, there is no correlation between default track points. To remedy this drawback, before using various classification methods, feature engineering is required to extract the relevant features between different trajectory points as much as possible. These feature extraction tend to be time consuming and labor intensive and do not extract the spatial correlation information between the trace points well, resulting in a lower Tian Lu segmentation accuracy.
(5) Segmentation method based on deep learning: the trajectory points are classified ("farmland" vs. "roads") using some common deep learning model (e.g., LSTM, CNN). These commonly used deep learning models are only applied to urban traffic track segmentation at present, and no related experiments are performed on agricultural machinery operation track Tian Lu segmentation. Moreover, these conventional deep learning models cannot simultaneously extract the spatial correlation information between different track points, which results in lower Tian Lu segmentation accuracy.
With the development of the modern agricultural process, the operation mode of individual planting is gradually changed into a collective planting mode, and a large amount of agricultural machine operation track data is generated in the agricultural production process. Therefore, the method for excavating the effective agricultural machinery operation track and calculating the farmland operation area has great significance for large-scale agricultural production management. The input amount of production data such as sowing, fertilizing, pesticide spraying and the like needs to estimate the farmland area, and the effective operation area of the agricultural machinery is calculated according to the operation track of the agricultural machinery, so that the method has great reference value for the field operation calculation time and charge of the agricultural machinery. The accurate farmland area can be accurately estimated by the accurate farmland dividing method, so that the manpower and material resources for calculating the farmland area can be reduced. The Beidou positioning global satellite positioning system can provide navigation and positioning information such as longitude, latitude, direction and the like of the agricultural machinery in real time, and is one of core technologies for supporting fine agricultural practice. By utilizing the Beidou positioning function, the longitude and latitude of the track points of the agricultural machinery at different moments can be measured, so that a data support is provided for the research of the Tian Lu segmentation method. The supervision of agricultural machinery track data and the mining of value information therein can promote agricultural modernization construction. Meanwhile, agricultural machinery operation track monitoring is also an important embodiment of a novel supervision mode of Internet and agricultural machinery operation.
Along with the rising and development of the internet of things technology, a data processing method based on location services is continuously appeared, and the precise agricultural field is the field based on the information technology. The agricultural machinery track data can be acquired by utilizing the navigation positioning terminal equipment, and the accuracy of farmland and road segmentation can be improved according to the information acquired after the data are processed.
Disclosure of Invention
The invention aims to provide a space-time diagram neural network-based agricultural machinery operation track Tian Lu segmentation method, which is characterized in that a space-time diagram neural network based on a time-space relation is adopted to segment an agricultural machinery effective operation track, a Beidou positioning acquisition module arranged on the agricultural machinery is utilized to acquire agricultural machinery track data, and the space-time diagram neural network is constructed by utilizing association information among different track points in the agricultural machinery track data so as to excavate more abundant track point characteristic information, so that automatic segmentation of farmlands and roads is realized on the agricultural machinery operation track; specifically, a space-time diagram is constructed aiming at the track data of one day of a certain farm machine, and different types of edges can be constructed by effectively utilizing the correlation between track points in the same farm; then, carrying out information propagation and feature extraction in the space-time diagram by adopting a diagram convolution network, and mining more abundant context information for each track point so as to improve the track point classification effect; the method specifically comprises three steps:
(1) And (5) extracting input characteristics: each row of data in the agricultural machinery track data contains the position, direction and speed information characteristics of the agricultural machinery at a certain moment; calculating longitude and latitude differences, acceleration and steering angles between two points by using the information, and enriching the characteristic dimensions; finally, each track point is represented by a vector formed by 7 features, and specifically comprises a longitude difference, a latitude difference, a speed, a direction difference, a steering angle, an acceleration and an acceleration change rate.
(2) And (3) constructing a space-time diagram: in order to fully utilize the space-time information contained in the track data, an algorithm for searching for adjacent points based on a time dimension and an algorithm for searching for adjacent points based on a space dimension are respectively designed, 7 track points related to space-time are found for each track point in each track, corresponding sides are constructed, and finally a space-time diagram of the whole track is obtained; specifically, the method can be divided into a plurality of steps of track point selection, time correlation edge construction, space correlation edge construction and autocorrelation edge construction.
(3) Convolution transformation of trace point features: carrying out multiple times of trace point characteristic information transfer by adopting a graph rolling network, and finally obtaining new characteristic representation of each trace point; because the association strength of each track point and the adjacent track points is different, each track point characteristic information transfer uses one aggregation layer to transfer information between track points; specifically, the adjacent matrix A of the input graph is transformed into a weighted matrix B, namely, in the weighted matrix B, the value representing the edge connection relation is not 1 any more, but a weight value, and in this way, the weighted average of all the neighbor track points can be realized; in addition, the weight values represent the association strength between track points, and the values are obtained by training a space-time diagram convolutional neural network; specifically, for the aggregation layer used for the 1 st trace point feature information transfer, its calculation is represented by the following function:
in the above, H (l) Is the input feature matrix of the 1 st aggregation layer, sigma is a nonlinear activation function,is a normalized matrix of the adjacency matrix A, W (l) Is a training weight.
The space-time diagram of the constructed track data is formed by a track point p at a moment t in the selected track data t As the reference point, find the nearest track point p in time at the previous time t-1 t-1 And a time sub-adjacent point p t-2 Similarly, p t+1 And p t+2 Respectively the locus points p t The track point and the time secondary adjacent point which are nearest in time at the next time t+1; secondly, searching for the track point p in the space distance from all track points except the track point of the nearest adjacent track point and the track point of the next adjacent track point t Nearest track point p s1 Secondary distance neighbor point p s2 。
The beneficial effects of the invention are as follows
(1) The first point of the invention has the outstanding effect: aiming at the problem that the traditional deep learning method in the agricultural machinery track data is poor in recognition effect, the invention provides a space-time diagram convolutional neural network for classifying the tracks of the track points, and the accuracy of Tian Lu segmentation can be integrally improved.
(2) The second point of the invention has outstanding effect: aiming at the space-time correlation between track points in the same farmland, the technology provides a space-time based two-dimensional composition mode, which can effectively improve the classification effect of the graph convolutional neural network.
(3) The third outstanding effect of the invention: a large number of experimental analyses show that the space-time diagram convolutional neural network is superior to the traditional road segmentation method not only in segmentation accuracy but also in model reliability.
Drawings
FIG. 1 is a sub-graph of input based on temporal and spatial dimensions.
Fig. 2 is a space-time based graph neural network convolution process.
Fig. 3 is an experimental flow chart.
Detailed Description
The invention provides a space-time diagram neural network-based agricultural machinery operation track Tian Lu segmentation method, which is characterized in that a space-time diagram neural network based on a time-space relation is adopted to segment an agricultural machinery effective operation track, the Beidou positioning acquisition module arranged on the agricultural machinery is utilized to acquire agricultural machinery track data, and the space-time diagram neural network is constructed by utilizing the association information among different track points in the agricultural machinery track data so as to excavate more abundant track point characteristic information, so that the agricultural machinery operation track is automatically segmented into farmlands and roads; specifically, a space-time diagram is constructed aiming at the track data of one day of a certain farm machine, and different types of edges can be constructed by effectively utilizing the correlation between track points in the same farm; then, carrying out information propagation and feature extraction in the space-time diagram by adopting a diagram convolution network, and mining more abundant context information for each track point so as to improve the track point classification effect; the invention will be further described with reference to the drawings and examples. The method comprises three steps:
(1) And (5) extracting input characteristics: each row of data in the agricultural machinery track data contains the position, direction and speed information characteristics of the agricultural machinery at a certain moment; calculating longitude and latitude differences, acceleration and steering angles between two points by using the information, and enriching the characteristic dimensions; finally, each track point is represented by a vector formed by 7 features, and specifically comprises a longitude difference, a latitude difference, a speed, a direction difference, a steering angle, an acceleration and an acceleration change rate.
(2) And (3) constructing a space-time diagram: in order to fully utilize the space-time information contained in the track data, an algorithm for searching for adjacent points based on a time dimension and an algorithm for searching for adjacent points based on a space dimension are respectively designed, 7 track points related to the space-time are found for each track point in each track, corresponding sides are constructed, and finally a space-time diagram (shown in figure 1) of the whole track is obtained; specifically, the method can be divided into a plurality of steps of track point selection, time correlation edge construction, space correlation edge construction and autocorrelation edge construction.
(3) Convolution transformation of trace point features: carrying out multiple times of trace point characteristic information transfer by adopting a graph rolling network (shown in figure 2), and finally obtaining new characteristic representation of each trace point; because the association strength of each track point and the adjacent track points is different, each track point characteristic information transfer uses one aggregation layer to transfer information between track points; specifically, the adjacent matrix A of the input graph is transformed into a weighted matrix B, namely, in the weighted matrix B, the value representing the edge connection relation is not 1 any more, but a weight value, and in this way, the weighted average of all the neighbor track points can be realized; in addition, the weight values represent the association strength between track points, and the values are obtained by training a space-time diagram convolutional neural network; for the aggregation layer used for the 1 st trace point feature information transfer, its calculation is represented by the following function:
in the above, H (l) Is the input feature matrix of the 1 st aggregation layer, sigma is a nonlinear activation function,is a normalized matrix of the adjacency matrix A, W (l) Is a training weight.
The space-time diagram of the constructed track data is formed by a track point p at a moment t in the selected track data t As the reference point, find the nearest track point p in time at the previous time t-1 t-1 And a time sub-adjacent point p t-2 Similarly, p t+1 And p t+2 Respectively the locus points p t The track point and the time secondary adjacent point which are nearest in time at the next time t+1; secondly, searching from all track points except the track point of the nearest time adjacent track point and the track point of the next time adjacent pointFind the point p of the track on the space distance t Nearest track point p s1 Secondary distance neighbor point p s2 。
Examples
The experimental data used in this example were all from wheat and rice harvesters from Wobbe farm, and there were 150 wheat harvester job trace data and 100 rice harvester job trace data in total. The positioning precision of the track points in the tracks is poor (about 2-5 meters), the operation sites are distributed in different provinces in China, and the operation conditions have extremely large differences, so that the traditional field road segmentation method has poor processing effect. The embodiment comprises the following steps:
step one, data acquisition, namely acquiring the daily operation track of the agricultural machinery by using a Beidou positioning acquisition module arranged on the agricultural machinery, wherein the daily track of each agricultural machinery is track data;
step two, data cleaning, in the track data acquisition process, sampling error conditions such as resampling, static track and the like often occur, and in order to avoid the influence of sampling errors on a subsequent classification method, the track data needs to be cleaned correspondingly, and the method specifically comprises the following steps:
1) Resampling type: the point with the time interval of 0S between the two points is cleaned, and the reserved point is the first point.
2) Type of rest trajectory: the point reserved point with the same longitude and latitude and the speed of 0 is still the first point after cleaning.
3) Repeat point type: and cleaning out continuous points which have the same longitude and latitude, the same speed and are not 0, wherein the reserved point is the first one.
4) Static drift type: and cleaning out continuous points with different longitudes and latitudes and 0 speed, wherein the reserved point is the first one.
5) Longitude and latitude anomaly type: when the latitude and longitude range exceeds the range of China, the condition that the acquired point is abnormal in latitude and longitude needs to be cleaned, and the cleaning mode is to delete the acquired point directly.
And thirdly, extracting input characteristics, namely respectively calculating attribute characteristics such as longitude difference, latitude difference, direction difference and the like of the track point at the current moment and the track point at the last moment for each track point in each track after data cleaning, so that each node is represented by a 7-dimensional vector.
And fourthly, constructing a space-time diagram, respectively designing an algorithm for searching for adjacent points based on a time dimension and for searching for adjacent points based on a space dimension in order to fully utilize space-time information contained in track data, finding 7 track points which are related in time and space for each track point of each track, constructing corresponding edges, and finally obtaining the space-time diagram of the whole track.
Specifically, a graph G= (V; E; R) is constructed for one-day operation track of a certain farm machine, wherein the node p j E V is the locus point, r (p i ,p j ) E is the connection trace point p i And p j Is R e R.
Node selection: the nodes in the graph are the trace points in the trace. As shown in fig. 2, a track has N track points, and there are N track points in the corresponding space-time diagram.
Construction of time-dependent edges: at a locus point p at a certain time t t For example, find its immediately preceding nearest track point p from the time dimension t-1 And the next adjacent track point p t-2 And then the nearest adjacent track point p t+1 And the next adjacent track point p t+2 The corresponding time dependent edges are constructed as shown in fig. 1.
Construction of spatially dependent edges: at the point of removal of the trace p t-2 、p t-1 、p t+1 、p t+2 In the subsequent track, a reference track point p is searched from the space dimension t Is the nearest neighbor distance track point p s1 And a secondary distance adjacent point locus point p s2 Corresponding spatially dependent edges are constructed as shown in fig. 1.
Construction of autocorrelation edges: construction of a track Point p t As shown in fig. 1.
And fifthly, carrying out convolution transformation on the track point characteristics, and carrying out multiple track point characteristic information transfer on the input space-time diagram by adopting a graph convolution method to finally obtain new characteristic representation of each track point. Because each track point has different association strength with the adjacent track points, each track point characteristic information transfer uses an aggregation layer to carry out convolution transformation between track points. Specifically, the adjacency matrix a of the input graph is transformed into a weighted matrix B, that is, in B, the values representing the edge connection relations are not all 1, but a weighted value, in this way a weighted average of all neighboring track points can be achieved. In addition, the weight values represent the association strength between the track points, and the values are obtained by training a space-time diagram convolutional neural network.
For the aggregation layer used for the 1 st trace point feature information transfer, its calculation is represented by the following function:
in the above, H (l) Is the input feature matrix of the 1 st aggregation layer, sigma is a nonlinear activation function,is a normalized matrix of the adjacency matrix A, W (l) Is a training weight.
And step six, classifying the track points, namely classifying the track points by adopting a linear network, and predicting whether the label of each track point is a farmland or a road.
Experimental results
Based on the above conception, a trace data space-time diagram was designed for the selected data, experiments were performed using a space-time diagram convolutional neural network (GCN) for the two data sets of wheat and rice, respectively, and the final experimental results are shown in tables 1 and 2. When the wheat harvesting track data are subjected to experiments, 150 pieces of track data are randomly divided into a training set, a verification set and a test set according to the proportion of 8:1:1. When the rice harvesting track data are subjected to experiments, 150 track data are randomly divided into a training set, a verification set and a test set according to the proportion of 8:1:1.
As can be seen from a comparison of table 1 and table 2, the Tian Lu segmentation effect of the rice harvesting trajectory is generally better than the Tian Lu segmentation effect of the wheat harvesting trajectory. For example, for the GCN model, the F1-score was 84.22% and 75.77%, respectively. This is due in large part to data imbalance problems. Specifically, in the wheat harvesting trace data, the ratio of Tian Lu trace points as shown in table 1 was 1:4; in the rice harvesting trace data, the ratio of Tian Lu trace points shown in Table 2 was 1:1.4.
Table 1 experimental effects of wheat harvester trajectory data
Farm land | Road | Average of | |
Accuracy rate of | 89.72 | 75.41 | 82.57 |
Recall rate of recall | 96.70 | 47.72 | 72.21 |
F1-score | 93.08 | 58.45 | 75.77 |
Table 2 experimental results of trajectory data of rice harvester
Farm land | Road | Average of | |
Accuracy rate of | 84.81 | 88.53 | 86.67 |
Recall rate of recall | 94.53 | 71.37 | 82.95 |
F1-score | 89.41 | 79.03 | 84.22 |
In addition, three common track segmentation methods, namely Random Forest (RF), decision Tree (DT) and long-short-term memory network (LSTM), are adopted, and comparison experiments are carried out on rice data and wheat data as a baseline model, and the final experimental comparison analysis results are shown in tables 3 and 4. The GCN-based agricultural work trajectory Tian Lu segmentation model performed best on these 2 data sets compared to the baseline model, indicating the reliability of the space-time convolutional neural network.
Table 3 experimental comparative analysis results of wheat harvester trajectory data
LSTM | DT | RF | GCN | |
Accuracy rate of | 79.73 | 72.12 | 75.94 | 82.57 |
Recall rate of recall | 66.18 | 53.27 | 53.82 | 72.21 |
F1-score | 69.71 | 52.01 | 52.91 | 75.77 |
Table 4 results of experimental comparative analysis of trajectory data of rice harvester
LSTM | DT | RF | GCN | |
Accuracy rate of | 80.55 | 71.13 | 73.60 | 86.67 |
Recall rate of recall | 76.42 | 69.29 | 71.77 | 82.95 |
F1-score | 77.59 | 69.85 | 72.38 | 84.22 |
In summary, the traditional classification and clustering algorithm has higher degree of dependence on the dimension and attribute of training data, and has poorer segmentation effect on the condition of wider farmland distribution area and fuzzy farmland road boundary. According to the invention, the Beidou positioning and acquisition module arranged on the agricultural machine is utilized to acquire the agricultural machine track data, a time-space diagram of the track data is constructed, the isolated track data can be associated based on the characteristic information of time and space, so that the excavated track data information is more abundant, and the neural network based on the time and space relation diagram is adopted to analyze the effective operation track of the agricultural machine. The problems of low farmland classification accuracy, incorrect classification of farmland road boundaries and the like of the traditional algorithm are solved, and reliable farmland and road segmentation of the agricultural machinery operation track is realized.
Claims (2)
1. The utility model provides a space-time diagram neural network-based agricultural machinery operation track Tian Lu segmentation method, which is characterized in that a space-time diagram neural network based on a time-space relation is adopted to segment an agricultural machinery effective operation track, the Beidou positioning acquisition module arranged on the agricultural machinery is utilized to acquire agricultural machinery track data, and the space-time diagram neural network is constructed by utilizing the association information among different track points in the agricultural machinery track data so as to excavate richer track point characteristic information, thereby realizing automatic segmentation of farmlands and roads for the agricultural machinery operation track; specifically, a space-time diagram is constructed aiming at the track data of one day of a certain farm machine, and different types of edges can be constructed by effectively utilizing the correlation between track points in the same farm; then, carrying out information propagation and feature extraction in the space-time diagram by adopting a diagram convolution network, and mining more abundant context information for each track point so as to improve the track point classification effect; the method specifically comprises three steps:
(1) And (5) extracting input characteristics: each row of data in the agricultural machinery track data contains the position, direction and speed information characteristics of the agricultural machinery at a certain moment; calculating longitude and latitude differences, acceleration and steering angles between two points by using the information, and enriching the characteristic dimensions; finally, each track point is represented by a vector consisting of 7 characteristics including longitude difference, latitude difference, speed, direction difference, steering angle, acceleration and acceleration change rate;
(2) And (3) constructing a space-time diagram: in order to fully utilize the space-time information contained in the track data, an algorithm for searching for adjacent points based on a time dimension and an algorithm for searching for adjacent points based on a space dimension are respectively designed, 7 track points related to space-time are found for each track point in each track, corresponding sides are constructed, and finally a space-time diagram of the whole track is obtained; specifically, the method comprises the steps of track point selection, time correlation edge construction, space correlation edge construction and autocorrelation edge construction;
(3) Convolution transformation of trace point features: carrying out multiple times of trace point characteristic information transfer by adopting a graph rolling network to obtain new characteristic representation of each trace point; because the association strength of each track point and the adjacent track points is different, each track point characteristic information transfer uses one aggregation layer to transfer information between track points; specifically, the adjacent matrix A of the input graph is transformed into a weighted matrix B, namely, in the weighted matrix B, the value representing the edge connection relation is not 1 any more, but a weight value, and in this way, the weighted average of all the neighbor track points can be realized; in addition, the weight values represent the association strength between nodes, and the values are obtained by training a space-time diagram convolutional neural network;
for the aggregation layer used for the first trace point information transfer, its calculation is represented by the following function:
in the above, H (l) Is the input feature matrix of the first aggregation layer, sigma is a nonlinear activation function,is a normalized matrix of the adjacency matrix A, W (l) Is a training weight.
2. The method for dividing an agricultural machinery operation track Tian Lu based on a space-time diagram neural network according to claim 1, wherein the constructed track data space-time diagram is formed by a track point p at a time t in selected track data t For reference point, finding the nearest track in time at the previous time t-1Trace point p t-1 And a time sub-adjacent point p t-2 Similarly, p t+1 And p t+2 Respectively the locus points p t The track point and the time secondary adjacent point which are nearest in time at the next time t+1; secondly, searching for the track point p in the space distance from all track points except the track point of the nearest adjacent track point and the track point of the next adjacent track point t Nearest track point p s1 Secondary distance neighbor point p s2 。
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210065003.8A CN114442623B (en) | 2022-01-20 | 2022-01-20 | Agricultural machinery operation track Tian Lu segmentation method based on space-time diagram neural network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210065003.8A CN114442623B (en) | 2022-01-20 | 2022-01-20 | Agricultural machinery operation track Tian Lu segmentation method based on space-time diagram neural network |
Publications (2)
Publication Number | Publication Date |
---|---|
CN114442623A CN114442623A (en) | 2022-05-06 |
CN114442623B true CN114442623B (en) | 2023-10-24 |
Family
ID=81367004
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202210065003.8A Active CN114442623B (en) | 2022-01-20 | 2022-01-20 | Agricultural machinery operation track Tian Lu segmentation method based on space-time diagram neural network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN114442623B (en) |
Families Citing this family (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114998744B (en) * | 2022-07-18 | 2022-10-25 | 中国农业大学 | Agricultural machinery track field dividing method and device based on motion and vision dual-feature fusion |
CN115881286B (en) * | 2023-02-21 | 2023-06-16 | 创意信息技术股份有限公司 | Epidemic prevention management scheduling system |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107145503A (en) * | 2017-03-20 | 2017-09-08 | 中国农业大学 | Remote supervision non-categorical relation extracting method and system based on word2vec |
CN111931905A (en) * | 2020-07-13 | 2020-11-13 | 江苏大学 | Graph convolution neural network model and vehicle track prediction method using same |
CN112905576A (en) * | 2021-03-02 | 2021-06-04 | 中国农业大学 | Method and system for determining farmland and road based on agricultural machinery operation track |
CN113111581A (en) * | 2021-04-09 | 2021-07-13 | 重庆邮电大学 | LSTM trajectory prediction method combining space-time factors and based on graph neural network |
CN113641773A (en) * | 2021-08-13 | 2021-11-12 | 中国农业大学 | Agricultural machinery behavior visualization marking method for driving track |
EP3937079A1 (en) * | 2020-07-09 | 2022-01-12 | Beijing Tusen Weilai Technology Co., Ltd. | Trajectory prediction method and device |
-
2022
- 2022-01-20 CN CN202210065003.8A patent/CN114442623B/en active Active
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107145503A (en) * | 2017-03-20 | 2017-09-08 | 中国农业大学 | Remote supervision non-categorical relation extracting method and system based on word2vec |
EP3937079A1 (en) * | 2020-07-09 | 2022-01-12 | Beijing Tusen Weilai Technology Co., Ltd. | Trajectory prediction method and device |
CN111931905A (en) * | 2020-07-13 | 2020-11-13 | 江苏大学 | Graph convolution neural network model and vehicle track prediction method using same |
CN112905576A (en) * | 2021-03-02 | 2021-06-04 | 中国农业大学 | Method and system for determining farmland and road based on agricultural machinery operation track |
CN113111581A (en) * | 2021-04-09 | 2021-07-13 | 重庆邮电大学 | LSTM trajectory prediction method combining space-time factors and based on graph neural network |
CN113641773A (en) * | 2021-08-13 | 2021-11-12 | 中国农业大学 | Agricultural machinery behavior visualization marking method for driving track |
Non-Patent Citations (1)
Title |
---|
多类农田障碍物卷积神经网络分类识别方法;薛金林;闫嘉;范博文;;农业机械学报(第S1期);42-48 * |
Also Published As
Publication number | Publication date |
---|---|
CN114442623A (en) | 2022-05-06 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108846832B (en) | Multi-temporal remote sensing image and GIS data based change detection method and system | |
CN114442623B (en) | Agricultural machinery operation track Tian Lu segmentation method based on space-time diagram neural network | |
CN112905576B (en) | Method and system for determining farmland and roads based on agricultural machinery operation track | |
CN108417033B (en) | Expressway traffic accident analysis and prediction method based on multi-dimensional factors | |
CN113010567B (en) | Agricultural machinery working area calculation method based on BDS/GPS positioning data space-time characteristics | |
CN111368736B (en) | Rice refined estimation method based on SAR and optical remote sensing data | |
CN112164062A (en) | Wasteland information extraction method and device based on remote sensing time sequence analysis | |
CN109063553A (en) | Field-crop growth defect area's remote sensing fast diagnosis method after a kind of land control | |
CN110929459A (en) | Site selection method for wind measuring tower of wind power plant in complex terrain | |
CN113011992A (en) | River basin agricultural non-point source pollution river entry coefficient measuring and calculating method based on standard data | |
CN102184423B (en) | Full-automatic method for precisely extracting regional impervious surface remote sensing information | |
CN113902580B (en) | Historical farmland distribution reconstruction method based on random forest model | |
CN111307164A (en) | Low-sampling-rate track map matching method | |
CN111144637A (en) | Regional power grid geological disaster forecasting model construction method based on machine learning | |
CN111523562A (en) | Commuting mode vehicle identification method based on license plate identification data | |
CN112734129B (en) | Air pollution space-time trend prediction method based on unsupervised restrictive optimization | |
CN111680888B (en) | Method for determining road network capacity based on RFID data | |
CN116631195B (en) | Regional abnormality detection method based on urban sub-region hot spot crossing mining | |
CN112052405B (en) | Passenger searching area recommendation method based on driver experience | |
CN112734120A (en) | Corn pest early warning method based on dynamic grid division | |
Mohamed et al. | Mapping techniques and intensity of soil sampling for precision farming | |
CN115310719A (en) | Farmland soil sampling scheme design method based on three-stage k-means | |
CN111476434B (en) | GIS-based soil heavy metal fractal dimension spatial variation analysis method | |
CN113011098B (en) | Lightning activity rule analysis display model and system for power transmission corridor area | |
CN106326264B (en) | Clustering apparatus and method |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |