Farmland land condition identification methods, devices, electronic equipment and storage media

Abstract

This invention provides a method, apparatus, electronic device, and storage medium for identifying farmland conditions. The method includes: acquiring static data of agricultural machinery, which characterizes the vehicle features of the agricultural machinery; acquiring operational status data of the agricultural machinery while it is driving in the farmland to be identified, which is time-series data with temporal correlation; and identifying the farmland conditions based on the static data, operational status data, and a neural network model to determine the target farmland condition category. The neural network model is trained on a sample dataset, which includes multiple sample data pairs and a farmland condition category label corresponding to each sample data pair. Each sample data pair includes sample static data of the agricultural machinery and sample operational status data of the agricultural machinery while it is driving in the sample farmland. Using this invention can improve the accuracy of farmland condition identification.

Description

Technical Field

[0001] This invention relates to the field of computer technology, and in particular to a method, apparatus, electronic device, and storage medium for identifying farmland conditions. Background Technology

[0002] Agricultural machinery automatic driving technology involves multiple disciplines. It primarily involves perceiving the environmental information surrounding the agricultural machinery and combining this with the vehicle's own positioning data to plan the most suitable path for movement and operation, thereby achieving autonomous movement and operation. Using this technology in farmland operations can effectively reduce agricultural labor intensity, improve work efficiency, and increase agricultural productivity. However, to better control the machinery and ensure it completes its tasks effectively, it is necessary to identify the conditions of the farmland where the machinery is operating.

[0003] In related technologies, when identifying the condition of farmland during agricultural machinery operations, one method is to directly observe the farmland on-site by personnel to obtain the farmland's condition; another method is to collect images of the farmland, extract features from the collected images, and then identify the farmland's condition based on the extracted features.

[0004] However, the above methods have drawbacks, such as being time-consuming and labor-intensive, or having low accuracy in identifying farmland conditions. Summary of the Invention

[0005] This invention provides a method, apparatus, electronic device, and storage medium for farmland condition identification, which addresses the shortcomings of existing technologies that rely on manual on-site identification of farmland conditions, leading to time and labor costs, or that rely on image features, resulting in low identification accuracy. The invention achieves comprehensive identification of farmland conditions by utilizing static data of agricultural machinery, operational status data showing changes in the movement characteristics of agricultural machinery while driving in the farmland, and neural network models. This saves time and labor costs in farmland condition identification while improving its accuracy.

[0006] This invention provides a method for identifying farmland conditions, comprising:

[0007] Obtain the static data corresponding to the agricultural machinery; the above static data is used to characterize the vehicle features of the agricultural machinery itself.

[0008] Acquire operational status data of agricultural machinery while it is operating in the farmland to be identified; the aforementioned operational status data is time-series data with time correlation.

[0009] Based on static data, operational status data, and a pre-set neural network model, the land conditions of the farmland to be identified are determined, and the target land condition category corresponding to the farmland to be identified is determined.

[0010] The aforementioned neural network model is trained based on a sample dataset, which includes multiple sample data pairs and the land condition category label of the sample farmland corresponding to each sample data pair. Each sample data pair includes sample static data of the agricultural machinery and sample operating status data of the agricultural machinery when it is driving in the sample farmland.

[0011] According to the farmland condition identification method provided by the present invention, the aforementioned neural network model includes a single-layer perceptron and a recurrent neural network model. The method identifies the farmland condition to be identified based on static data, operational status data, and a preset neural network model, determining the target farmland condition category corresponding to the farmland to be identified, including:

[0012] Static data is input into a single-layer perceptron for feature extraction to determine the first feature corresponding to the static data.

[0013] The operational status data is input into a recurrent neural network model for feature extraction to determine the second feature corresponding to the operational status data; the second feature is time-dependent and related to the land conditions of the farmland to be identified.

[0014] Based on the first and second features, the land condition of the farmland to be identified is determined, and the target land condition category corresponding to the farmland to be identified is determined.

[0015] According to the farmland condition identification method provided by the present invention, the aforementioned neural network model further includes a first fully connected layer, a second fully connected layer, and an activation layer. The method identifies the farmland condition to be identified based on a first feature and a second feature, determining the target farmland condition category corresponding to the farmland to be identified, including:

[0016] The first feature and the second feature are input into the first fully connected layer for feature fusion processing to determine the first fused feature;

[0017] The first fusion feature is input into the second fully connected layer for feature fusion processing to determine the second fusion feature;

[0018] The second fusion feature is input into the activation layer for classification processing to determine the target land condition category corresponding to the farmland to be identified.

[0019] According to the farmland condition identification method provided by the present invention, the above-mentioned recurrent neural network model includes a long short-term memory network model, and the long short-term memory network model includes a multi-layer long short-term memory network.

[0020] According to a method for identifying farmland conditions provided by the present invention, the above-mentioned acquisition of the operating status data of agricultural machinery when it is traveling in the farmland to be identified includes:

[0021] Acquire the initial operating status data of the agricultural machinery when it is driving in the farmland to be identified; the initial operating status data includes multiple initial sub-data sorted by time;

[0022] Data cleaning and normalization are performed on multiple initial sub-data of the initial running state data to obtain multiple preprocessed sub-data corresponding to the initial running state data;

[0023] Filter the preprocessed sub-data of the initial running state data to determine the filtered sub-data corresponding to the initial running state data.

[0024] The multiple sub-data points after filtering the initial operating status data are combined to determine the operating status data of the agricultural machinery when it is driving in the farmland to be identified.

[0025] According to the farmland condition identification method provided by the present invention, the method further includes:

[0026] Obtain the sample dataset; the sample dataset includes multiple sample data pairs and the land condition category label of the sample farmland corresponding to each sample data pair. Each sample data pair includes sample static data of agricultural machinery and sample operating status data of agricultural machinery when it is driving in the sample farmland.

[0027] For each sample data pair, the predicted land condition category corresponding to the sample data pair is determined based on the sample static data and sample operating status data of agricultural machinery in the sample data pair, as well as the initial neural network model.

[0028] Based on the predicted land condition categories and corresponding land condition category labels of the sample data pairs, a regularization method is used to train the initial neural network model to determine the neural network model.

[0029] According to the farmland condition identification method provided by the present invention, the method further includes:

[0030] Based on the target land condition category corresponding to the farmland to be identified, the control parameters for the agricultural machinery to travel in the farmland to be identified are dynamically adjusted; the aforementioned control parameters include the agricultural machinery's speed and / or direction of travel.

[0031] The present invention also provides a farmland condition identification device, comprising the following modules:

[0032] The static data acquisition module is used to acquire the static data corresponding to the agricultural machinery; the aforementioned static data is used to characterize the vehicle features of the agricultural machinery itself.

[0033] The operation status data acquisition module is used to acquire the operation status data of agricultural machinery when it is driving in the farmland to be identified; the above operation status data is time series data with time correlation.

[0034] The land condition identification module is used to identify the land condition of the farmland to be identified based on static data, operational status data, and a preset neural network model, and to determine the target land condition category corresponding to the farmland to be identified. The neural network model is trained based on a sample dataset, which includes multiple sample data pairs and the land condition category label of the sample farmland corresponding to each sample data pair. Each sample data pair includes sample static data of the agricultural machinery and sample operational status data of the agricultural machinery when it is driving in the sample farmland.

[0035] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the farmland condition identification method as described above.

[0036] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the farmland condition identification method as described above.

[0037] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements a farmland condition identification method as described above.

[0038] The farmland condition identification method, device, electronic device, and storage medium provided by this invention acquire static data characterizing the vehicle features of the agricultural machinery itself, and acquire operational status data of the agricultural machinery while driving in the farmland to be identified. Based on the static data, operational status data, and a pre-trained neural network model, the method identifies the farmland condition and determines the target farmland condition category. The operational status data of the agricultural machinery is time-series data with time correlation. The neural network model is trained on a sample dataset including multiple sample data pairs and the farmland condition category label corresponding to each sample data pair. Each sample data pair includes sample static data of the agricultural machinery and sample operational status data of the agricultural machinery while driving in the sample farmland. This method identifies farmland conditions using multimodal data, including static data and operational data of agricultural machinery in the field. This provides a richer dataset that accurately reflects the farmland's conditions on the machinery. Furthermore, by combining this data with a pre-trained neural network model, the identification is more accurate, resulting in higher precision. Additionally, this method eliminates the need for manual observation of farmland conditions, instead automatically collecting multimodal data from agricultural machinery and combining it with a neural network model for identification, thus saving time and labor costs. Attached Figure Description

[0039] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0040] Figure 1 This is one of the flowcharts illustrating the farmland condition identification method provided by the present invention.

[0041] Figure 2 This is the second flowchart of the farmland condition identification method provided by the present invention.

[0042] Figure 3 This is a schematic diagram of the neural network model architecture provided by the present invention.

[0043] Figure 4 This is the third flowchart of the farmland condition identification method provided by the present invention.

[0044] Figure 5 This is a schematic diagram of the farmland condition identification device provided by the present invention.

[0045] Figure 6 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0046] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0047] Currently, when identifying the condition of farmland under agricultural machinery operation, one method is to directly observe the farmland on-site, i.e., through manual surveys to obtain information about the farmland's condition. However, this method is time-consuming and labor-intensive, and the identification results are easily affected by subjective factors. Another method is to acquire images of the farmland, extract features from the acquired images, and then identify the farmland's condition based on the extracted features. However, this method has low accuracy in identifying farmland conditions. Therefore, embodiments of the present invention provide a method, apparatus, electronic device, and storage medium for farmland condition identification, which can solve the above-mentioned technical problems.

[0048] The following is combined with Figures 1-4 The present invention describes a method for identifying farmland conditions.

[0049] It should be noted that the execution subject of this invention embodiment can be a farmland condition identification device, an electronic device, or other devices or equipment. The following embodiments will use an electronic device as the execution subject for illustration. The electronic device can be a terminal or server outside the agricultural machinery, and both can interact with the processor / controller and other electronic devices in the agricultural machinery for data / instruction.

[0050] Figure 1 This is one of the flowcharts illustrating the farmland condition identification method provided by the present invention, such as... Figure 1 As shown, the method includes the following steps:

[0051] Step 102: Obtain the static data corresponding to the agricultural machinery; the above static data is used to characterize the vehicle features of the agricultural machinery itself.

[0052] Agricultural machinery refers to agricultural vehicles, which are generally used in farmland, and may include harvesters, tillers, etc.

[0053] Static data of agricultural machinery characterizes its vehicle features. This static data is generally fixed after the machinery is stationary and does not change when the machinery is in normal working order. This static data may include, for example, the wheelbase, minimum turning radius, and vehicle model. The vehicle model can be an Ackermann model or a differential model, etc. Furthermore, the types and quantities of static data provided here are merely examples and not specific limitations.

[0054] Static data for agricultural machinery can be obtained from the machinery's factory parameters; alternatively, it can be obtained by creating a 3D model of the machinery and simulating the static data; it can also be obtained by manually measuring the vehicle characteristics of the machinery; or other methods can be used. No specific limitations are made here, as long as the static data of the agricultural machinery can be obtained.

[0055] Step 104: Obtain the operating status data of the agricultural machinery while it is driving in the farmland to be identified; the above operating status data is time series data with time correlation.

[0056] The farmland requiring terrain identification can be designated as "farmland to be identified," and this farmland can be predetermined. After identifying the farmland to be identified, the agricultural machinery can be controlled to travel / operate within it for a preset time. Then, motion characteristic data of the machinery at each moment within this preset time is acquired. Combining these motion characteristic data yields the operational status data of the agricultural machinery within the preset time. This preset time is a relatively short period, such as 5 minutes or 10 minutes.

[0057] Understandably, agricultural machinery operation status data refers to the motion characteristic data of agricultural machinery traveling in the farmland to be identified at various moments within a preset time period. This data reflects the operational status of the agricultural machinery while traveling in the farmland, i.e., the changes in motion characteristics. This operational status data of agricultural machinery has time correlation and can also be called time series data or dynamic change data. This time series data may include data such as time series length, batch size, and feature dimensions. The time series length can be the duration of the collected operational status data, i.e., the preset time; the batch size can be the number of motion features included in the operational status data collected at each moment; and the feature dimensions can be the number of dimensions used to represent each motion feature.

[0058] The operational status data of agricultural machinery can include operational status data with one feature or multiple features. These features can include, for example, the coordinates, angular velocity, acceleration, and heading angle of the agricultural machinery when it is traveling in the farmland to be identified.

[0059] In addition, various vehicle condition sensors can be installed on agricultural machinery to collect motion characteristic data of the agricultural machinery at various times within a preset time when the agricultural machinery is traveling in the farmland to be identified. In this way, the operating status data of the agricultural machinery when traveling in the farmland to be identified within a preset time can be obtained.

[0060] Step 106: Based on static data, operational status data, and a preset neural network model, identify the land conditions of the farmland to be identified and determine the target land condition category corresponding to the farmland to be identified.

[0061] To identify farmland conditions more quickly and accurately, this embodiment proposes using a neural network model. This model can be composed of a perceptual network, a recurrent neural network, or other types of neural networks; its specific type and architecture are not specifically limited here. The main function of this neural network model is to identify the farmland conditions based on the static and operational data of the agricultural machinery obtained above. It can be pre-trained. The training process involves training the neural network model on a sample dataset. This dataset includes multiple sample data pairs and a corresponding farmland condition category label for each pair. Each sample data pair includes sample static data of the agricultural machinery and sample operational status data of the machinery while it is operating in the sample farmland. In other words, the neural network model is trained on a large sample dataset labeled with farmland condition category labels. This trained neural network model can accurately identify the farmland conditions.

[0062] Specifically, after the neural network model is trained and the static and operational status data of the agricultural machinery are obtained, the static and operational status data can be directly input into the neural network model for land condition identification processing to obtain the target land condition category corresponding to the farmland to be identified. Alternatively, the static and operational status data of the agricultural machinery can be preprocessed before being input into the neural network model for land condition identification processing to obtain the target land condition category corresponding to the farmland to be identified. Furthermore, during land condition identification processing in the neural network model, feature extraction, feature fusion, and classification can be performed on the static and operational status data of the agricultural machinery to ultimately obtain the target land condition category of the farmland to be identified.

[0063] Furthermore, the land condition category can refer to the type of farmland, such as hard soil, soft soil, sandy soil, or paddy field. Of course, the land condition category can also be other attributes of the farmland. Identifying the land condition category of farmland can help agricultural machinery operate more effectively in the fields.

[0064] Based on the target land condition category corresponding to the farmland to be identified above, as an optional embodiment, the control parameters of the agricultural machinery driving in the farmland to be identified can be dynamically adjusted according to the target land condition category corresponding to the farmland to be identified; the control parameters include the driving speed and / or driving direction of the agricultural machinery.

[0065] In other words, after obtaining the target terrain category corresponding to the farmland to be identified, the speed or direction of the agricultural machinery can be dynamically adjusted based on the target terrain category. For example, if the target terrain category is hard ground, the speed of the agricultural machinery can be slightly increased to ensure that it can quickly operate along the planned path, and the direction of the agricultural machinery will not change much, requiring only minor adjustments. On the other hand, if the target terrain category is soft ground, the speed of the agricultural machinery can be slightly decreased to ensure that it can travel along the planned path. The direction of the agricultural machinery may change frequently, so the direction of travel can be dynamically adjusted multiple times to ensure that the agricultural machinery operates according to the planned path and direction.

[0066] By dynamically adjusting the control parameters of agricultural machinery as it travels in the identified farmland by classifying the target terrain, precise control of the machinery can be achieved. This improves the working / operational accuracy of the machinery and reduces the cost of manual intervention, enabling automated, efficient, and precise operation of the machinery in the farmland.

[0067] In this embodiment, static data characterizing the vehicle features of the agricultural machinery and operational status data of the agricultural machinery while it is driving in the farmland to be identified are acquired. Based on the static data, operational status data, and a pre-trained neural network model, the terrain conditions of the farmland to be identified are determined, and the target terrain condition category corresponding to the farmland is identified. The operational status data of the agricultural machinery is time-series data with time correlation. The neural network model is trained on a sample dataset that includes multiple sample data pairs and the terrain condition category label of the sample farmland corresponding to each sample data pair. Each sample data pair includes sample static data of the agricultural machinery and sample operational status data of the agricultural machinery while it is driving in the sample farmland. This method identifies farmland conditions using multimodal data, including static data and operational data of agricultural machinery in the field. This provides a richer dataset that accurately reflects the farmland's conditions on the machinery. Furthermore, by combining this data with a pre-trained neural network model, the identification is more accurate, resulting in higher precision. Additionally, this method eliminates the need for manual observation of farmland conditions, instead automatically collecting multimodal data from agricultural machinery and combining it with a neural network model for identification, thus saving time and labor costs.

[0068] In order to accurately identify the land condition categories of farmland, the embodiments of the present invention have made specific improvements to the architecture of the neural network model. Specifically, the aforementioned neural network model may include a single-layer perceptron and a recurrent neural network model. The following embodiments will explain how to perform land condition identification under this model architecture.

[0069] Figure 2 This is the second flowchart of the farmland condition identification method provided by the present invention, as shown below. Figure 2 As shown, step 106 above, "identifying the land condition of the farmland to be identified based on static data, operational status data, and a preset neural network model, and determining the target land condition category corresponding to the farmland to be identified," may include the following steps:

[0070] Step 202: Input the static data into a single-layer perceptron for feature extraction and determine the first feature corresponding to the static data.

[0071] In this step, see Figure 3 The schematic diagram of the neural network model architecture provided by the present invention is shown. The neural network model includes two parallel sub-networks (i.e., the neural network model has two model channels), one is a sub-network composed of a single-layer perceptron (SLP), and the other is a sub-network composed of a recurrent neural network model.

[0072] The single-layer perceptron (SLP) consists of an input layer and an output layer, which are directly connected. This allows for relatively rapid feature extraction and classification from the data. After obtaining the static data of the agricultural machinery, this data can be input into the single-layer perceptron in the neural network model for feature extraction, obtaining the features corresponding to the static data of the agricultural machinery, denoted as the first feature.

[0073] Step 204: Input the operating status data into the recurrent neural network model for feature extraction and determine the second feature corresponding to the operating status data; the second feature is time-related and related to the land conditions of the farmland to be identified.

[0074] As mentioned above, neural network models can also include recurrent neural network (RNN) models. RNN models are a type of recurrent neural network that takes sequence data as input, recursively moves in the direction of sequence evolution, and all nodes (or recurrent units) are connected in a chain. They can effectively process and remember the time dependencies in the data and generate more accurate output results based on the context information in the input sequence data.

[0075] Specifically, after obtaining the operational status data of the agricultural machinery traveling in the farmland to be identified within a preset time period, this operational status data can be input into a recurrent neural network model within the neural network model for feature extraction. This extracts the dynamically changing features of the agricultural machinery during operation, denoted as the second feature. This second feature, like the operational status data input into the recurrent neural network model, also exhibits temporal correlation and can reflect the terrain conditions of the farmland to be identified. For the same agricultural machinery, the operational status data differs when traveling in farmland with different terrain conditions. Therefore, by analyzing the operational status data of the agricultural machinery traveling in the farmland to be identified, the terrain conditions of the farmland can be more accurately identified.

[0076] In addition, as an optional embodiment, the recurrent neural network model described above may include a long short-term memory (LSTM) network model, and the LSTM network model may include a multi-layer LSTM network.

[0077] The LSTM model effectively preserves and updates long-term information by introducing cell states and gating mechanisms. It can learn long-term dependencies and selectively retain or discard information, thus effectively mitigating the vanishing and exploding gradient problems, thereby improving the accuracy of extracted features and subsequent recognition. Furthermore, the LSTM model in this embodiment can be constructed using a multi-layer LSTM network, with each layer being a bidirectional LSTM network. The specific number of layers is not limited; for example, it could be a two-layer or three-layer LSTM network. Feature information can be transferred between multiple LSTM networks, further improving the accuracy of extracted features and the final recognition accuracy.

[0078] Specifically, after obtaining the operating status data of the agricultural machinery traveling in the farmland to be identified within a preset time, the operating status data can be input into the multi-layer LSTM model in the neural network model for feature extraction to obtain the second feature of the agricultural machinery dynamically changing during operation.

[0079] Step 206: Based on the first feature and the second feature, identify the land condition of the farmland to be identified and determine the target land condition category corresponding to the farmland to be identified.

[0080] In this step, after obtaining the first feature corresponding to the static data of the agricultural machinery and the second feature corresponding to the operating status data, the two feature data can be combined to identify the land condition category of the farmland to be identified.

[0081] As an optional embodiment, the above-mentioned neural network model further includes a first fully connected layer, a second fully connected layer, and an activation layer. The process of identifying the land condition of the farmland to be identified through the first feature and the second feature may include: inputting the first feature and the second feature into the first fully connected layer for feature fusion processing to determine the first fused feature; inputting the first fused feature into the second fully connected layer for feature fusion processing to determine the second fused feature; and inputting the second fused feature into the activation layer for classification processing to determine the target land condition category corresponding to the farmland to be identified.

[0082] In other words, after obtaining the first and second features of the agricultural machinery, these features can be input into the first fully connected layer for feature fusion processing to obtain the initial fused features, denoted as the first fused feature. This first fused feature is then further input into the second fully connected layer for feature fusion processing again. During this second fusion process, the first and second features can be added again to continue the fusion, ultimately obtaining the fused features, denoted as the second fused feature. Next, the second fused feature can be input into the softmax activation layer (also known as the output layer). The softmax activation function is used to classify the second fused feature, outputting the classification result, which is the probability that the farmland to be identified belongs to each land condition category. Then, based on the probabilities of each land condition category, the land condition category with the highest probability is selected as the target land condition category corresponding to the farmland to be identified.

[0083] Furthermore, the aforementioned single-layer perceptron, first fully connected layer, and second fully connected layer can all be constructed using multiple dense networks. Dense networks can significantly reduce the number of parameters and improve feature utilization through feature reuse and dense connections, thereby enhancing feature extraction accuracy and improving land condition category recognition precision.

[0084] Furthermore, based on the aforementioned neural network model architecture, the training process of the neural network model may include the following steps:

[0085] Obtain the sample dataset; the sample dataset includes multiple sample data pairs and the land condition category label of the sample farmland corresponding to each sample data pair. Each sample data pair includes sample static data of agricultural machinery and sample operating status data of agricultural machinery when it is driving in the sample farmland.

[0086] For each sample data pair, the predicted land condition category corresponding to the sample data pair is determined based on the sample static data and sample operating status data of agricultural machinery in the sample data pair, as well as the initial neural network model.

[0087] Based on the predicted land condition categories and corresponding land condition category labels of the sample data pairs, a regularization method is used to train the initial neural network model to determine the neural network model.

[0088] This process involves pre-labeling the terrain of each sample farmland to obtain a terrain category label for each sample farmland. Then, it allows for the collection of sample operational status data of the same agricultural machinery operating in different sample farmlands within a preset time period, resulting in multiple sets of sample operational status data for the same machinery. Simultaneously, static data of the same agricultural machinery can be collected as sample static data. By binding a set of sample operational status data, sample static data, and the corresponding sample farmland terrain category label together, a set of sample data pairs and their terrain category labels can be obtained. By binding all sample operational status data with sample static data and the corresponding sample farmland terrain category labels, multiple sets of sample data pairs and their terrain category labels can be obtained.

[0089] Alternatively, sample operational status data and static data of different agricultural machines operating in different sample farmlands within a preset time period can be collected to obtain sample data pairs. Furthermore, the aforementioned sample operational status data can also include multiple sets of sample operational status data obtained after the same agricultural machine has been used multiple times in the same sample farmland. This can enrich the sample data and improve the accuracy of land condition identification.

[0090] Multiple sets of sample data pairs can be obtained through the above method. These pairs can then be combined to obtain a sample dataset. After obtaining the sample dataset, each sample data pair can be input into the initial neural network model for land condition identification. Specifically, the static data of each sample data pair can be input into a single-layer perceptron in the initial neural network model for feature extraction to obtain the first feature. The operational state data of each sample data pair can be input into a multi-layer LSTM network in the initial neural network model for feature extraction to obtain the second feature. Then, the first and second features are fused twice and input into the activation layer for classification processing to obtain the predicted land condition category corresponding to the sample data pair, i.e., the land condition category predicted by the initial neural network model. Next, the loss between the predicted land condition category and the land condition category label of the sample data pair can be calculated. This loss can be cross-entropy loss, etc. Then, based on the calculated loss, the parameters of the initial neural network model are adjusted using the Adam (Adaptive Momentum) optimization algorithm to train the initial neural network model.

[0091] In addition, to avoid overfitting during the initial neural network model training process, regularization methods can be introduced, i.e., regularization methods are used to train the initial neural network model. These regularization methods can be Dropout (random deactivation) or L2 regularization, etc.

[0092] By training the initial neural network model using the aforementioned loss function, Adam optimization algorithm, and regularization method, a well-trained neural network model can be obtained. This type of training avoids overfitting, resulting in higher recognition accuracy.

[0093] In this embodiment, a single-layer perceptron in the neural network model extracts features from the static data of the agricultural machinery, and a recurrent neural network model extracts features from the operational status data of the agricultural machinery. Based on the two features extracted from these two channels, farmland conditions are identified. This approach allows for rapid and accurate extraction of multimodal data from the agricultural machinery using both the single-layer perceptron and the recurrent neural network model, thereby improving the accuracy and efficiency of subsequent land condition identification. Furthermore, employing a multi-layer long short-term memory network to extract features from the operational status data further enhances the extraction of farmland condition-related features, further improving the accuracy of farmland condition identification. Moreover, by performing two feature fusion processes on the extracted multimodal features of the agricultural machinery before classification, even better farmland condition features are obtained, further improving the accuracy of farmland condition identification.

[0094] The following examples illustrate the process of acquiring operational status data of agricultural machinery while it is traveling in the farmland to be identified within a preset time period.

[0095] Figure 4 This is the third flowchart of the farmland condition identification method provided by the present invention, as shown below. Figure 4 As shown, step 104 above, "obtaining the operating status data of the agricultural machinery while it is operating in the farmland to be identified," may include the following steps:

[0096] Step 402: Obtain the initial operating status data of the agricultural machinery when it is driving in the farmland to be identified; the initial operating status data includes multiple initial sub-data sorted by time.

[0097] As mentioned in the above embodiments, at various times within a preset time period, the operating status data of the agricultural machinery while it is driving in the farmland to be identified can be collected by the sensors on the agricultural machinery. The operating status data collected at each time is obtained, and these data are sorted according to the collection time. The data collected at each time is recorded as the initial sub-data, so that multiple initial sub-data sorted by time can be obtained. Then, the multiple initial sub-data sorted by time can be recorded as the initial operating status data of the agricultural machinery.

[0098] Furthermore, the initial motion state data of agricultural machinery can be a single motion state data or multiple operational state data. If the initial operational state data includes multiple initial operational state data, then each initial operational state data can include multiple initial sub-data ordered by time.

[0099] For example, if the initial operating status data includes multiple initial operating status data, these multiple initial operating status data may include, for example, the coordinates, linear velocity, angular velocity, acceleration (including acceleration along the x, y, and z axes), roll angle, absolute value of the front error, absolute value of the rear error, heading angle, and heading error of the agricultural machinery at various times. The coordinates, linear velocity, angular velocity, acceleration, roll angle, and heading angle data can all be obtained by collecting data from corresponding sensors installed on the agricultural machinery or calculated from the collected data. The absolute values ​​of the front error, rear error, and heading error data can be obtained by calculating the position or angle of the corresponding areas collected by sensors installed in the front, rear, and body areas of the machinery.

[0100] Step 404: Perform data cleaning and normalization on multiple initial sub-data of the initial running state data to obtain multiple preprocessed sub-data corresponding to the initial running state data.

[0101] Data cleaning of multiple initial sub-data sets can include removing outlier data and other abnormal data from the multiple initial sub-data sets.

[0102] Specifically, for each type of initial running state data, after obtaining its corresponding multiple initial sub-data sorted by time, these initial sub-data can be cleaned first to obtain multiple cleaned sub-data. Then, the multiple cleaned sub-data can be normalized to unify the range of the multiple cleaned sub-data, forming data suitable for input to the neural network model, which is convenient for subsequent input into the neural network model for feature extraction. Here, the normalized sub-data can be obtained through normalization, or it can be referred to as the preprocessed sub-data.

[0103] Step 406: Filter the preprocessed sub-data of the initial running state data to determine the filtered sub-data corresponding to the initial running state data.

[0104] In this step, for each type of initial running state data, after obtaining its corresponding preprocessed sub-data, the preprocessed sub-data can be filtered to obtain filtered sub-data.

[0105] As an optional embodiment, a moving average filtering algorithm or a symmetric exponential moving average filtering algorithm can be used for filtering. Specifically, the average value of data within a specific time window can be selected to represent the data at that moment. This specific time window could be, for example, 10 seconds, 20 seconds, etc. For example, for each sub-data point at a given moment, the average value of sub-data points within 10 seconds starting from that moment can be selected as the final sub-data point for that moment. Optionally, the number of sub-data points after the above filtering process can be equal to or less than the number of sub-data points after preprocessing.

[0106] The above method allows for data cleaning, normalization, and filtering of multiple sub-data points for each initial operating state, ultimately yielding multiple filtered sub-data points corresponding to each initial operating state, thus improving the accuracy of the obtained operating state data.

[0107] Step 408: Combine multiple sub-data after filtering the initial operating status data to determine the operating status data of the agricultural machinery when it is driving in the farmland to be identified.

[0108] In this step, after obtaining multiple filtered sub-data corresponding to each initial operating state data, the multiple filtered sub-data of each initial operating state data can be treated as a time series data. In this way, the time series data corresponding to each initial operating state data can be obtained. Then, these time series data can be used as the operating state data of the agricultural machinery when it is driving in the farmland to be identified.

[0109] In this embodiment, by performing data cleaning, normalization, and filtering on multiple sub-data of the initial operating status data, the operating status data of the agricultural machinery when it is driving in the farmland to be identified is finally obtained. This multi-processing can improve the accuracy of the final obtained operating status data of the agricultural machinery, thereby improving the accuracy of the farmland conditions identified in the subsequent process.

[0110] The farmland condition identification device provided by the present invention is described below. The farmland condition identification device described below and the farmland condition identification method described above can be referred to in correspondence.

[0111] Figure 5 This is a schematic diagram of the farmland condition identification device provided by the present invention. (See attached diagram) Figure 5 As shown, the device may include:

[0112] The static data acquisition module 510 is used to acquire static data corresponding to the agricultural machinery; the aforementioned static data is used to characterize the vehicle features of the agricultural machinery itself.

[0113] The operation status data acquisition module 520 is used to acquire the operation status data of the agricultural machinery when it is driving in the farmland to be identified; the above operation status data is time series data with time correlation.

[0114] The land condition identification module 530 is used to identify the land condition of the farmland to be identified based on static data, operational status data and a preset neural network model, and determine the target land condition category corresponding to the farmland to be identified. The neural network model is trained based on a sample dataset, which includes multiple sample data pairs and the land condition category label of the sample farmland corresponding to each sample data pair. Each sample data pair includes sample static data of the agricultural machinery and sample operational status data of the agricultural machinery when it is driving in the sample farmland.

[0115] In some embodiments, the neural network model includes a single-layer perceptron and a recurrent neural network model. The land condition identification module 530 is specifically used to input static data into the single-layer perceptron for feature extraction to determine the first feature corresponding to the static data; input operational status data into the recurrent neural network model for feature extraction to determine the second feature corresponding to the operational status data; the second feature is time-dependent and related to the land condition of the farmland to be identified; and based on the first and second features, the land condition of the farmland to be identified is identified to determine the target land condition category corresponding to the farmland to be identified.

[0116] Optionally, the aforementioned neural network model further includes a first fully connected layer, a second fully connected layer, and an activation layer. The aforementioned land condition recognition module 530 is specifically used to input the first feature and the second feature into the first fully connected layer for feature fusion processing to determine the first fused feature; input the first fused feature into the second fully connected layer for feature fusion processing to determine the second fused feature; and input the second fused feature into the activation layer for classification processing to determine the target land condition category corresponding to the farmland to be identified.

[0117] Optionally, the above recurrent neural network model includes a long short-term memory network model, and the long short-term memory network model includes a multi-layer long short-term memory network.

[0118] In some embodiments, the above-mentioned operating status data acquisition module 520 is specifically used to acquire the initial operating status data of the agricultural machinery when it is driving in the farmland to be identified; the initial operating status data includes multiple initial sub-data ordered by time; the multiple initial sub-data of the initial operating status data are cleaned and normalized to obtain multiple pre-processed sub-data corresponding to the initial operating status data; the multiple pre-processed sub-data of the initial operating status data are filtered to determine the multiple filtered sub-data corresponding to the initial operating status data; the multiple filtered sub-data of the initial operating status data are combined to determine the operating status data of the agricultural machinery when it is driving in the farmland to be identified.

[0119] In some embodiments, the above-described apparatus further includes:

[0120] The model training module is used to acquire a sample dataset. This dataset includes multiple sample data pairs and a land condition category label for each sample farmland. Each sample data pair includes static data of the agricultural machinery and operational status data of the machinery while it is operating in the sample farmland. For each sample data pair, the predicted land condition category is determined based on the static and operational data of the agricultural machinery, the initial neural network model, and the predicted land condition category. Based on the predicted land condition category and the corresponding land condition category label, a regularization method is used to train the initial neural network model, thus determining the neural network model.

[0121] In some embodiments, the above-described apparatus further includes:

[0122] The adjustment module is used to dynamically adjust the control parameters of the agricultural machinery when it travels in the farmland to be identified, based on the target land condition category corresponding to the farmland to be identified; the control parameters include the speed and / or direction of travel of the agricultural machinery.

[0123] It should be noted that the apparatus provided in this embodiment of the invention can implement all the method steps implemented in the above method embodiment and can achieve the same technical effect. Therefore, the parts and beneficial effects that are the same as those in the method embodiment will not be described in detail here.

[0124] Figure 6 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 6As shown, the electronic device may include: a processor 610, a communication interface 620, a memory 630, and a communication bus 640, wherein the processor 610, the communication interface 620, and the memory 630 communicate with each other through the communication bus 640. The processor 610 can call logical instructions in the memory 630 to execute a farmland condition identification method, which includes: acquiring static data corresponding to the agricultural machinery; the static data being used to characterize the vehicle features of the agricultural machinery; acquiring the operating status data of the agricultural machinery while it is driving in the farmland to be identified; the operating status data being time-series data with time correlation; and identifying the farmland condition to be identified based on the static data, the operating status data, and a preset neural network model, thereby determining the target farmland condition category corresponding to the farmland to be identified; wherein the neural network model is trained based on a sample dataset, the sample dataset including multiple sample data pairs and the farmland condition category label corresponding to each sample data pair, and each sample data pair including sample static data of the agricultural machinery and sample operating status data of the agricultural machinery while it is driving in the sample farmland.

[0125] Furthermore, the logical instructions in the aforementioned memory 630 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0126] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the farmland condition identification method provided by the above methods. The method includes: acquiring static data corresponding to agricultural machinery; the static data being used to characterize the vehicle features of the agricultural machinery itself; acquiring the operating status data of the agricultural machinery while it is driving in the farmland to be identified; the operating status data being time-series data with time correlation; identifying the farmland condition to be identified based on the static data, the operating status data, and a preset neural network model, and determining the target farmland condition category corresponding to the farmland to be identified; wherein the neural network model is trained based on a sample dataset, the sample dataset including multiple sample data pairs and the farmland condition category label corresponding to each sample data pair, each sample data pair including sample static data of the agricultural machinery and sample operating status data of the agricultural machinery while it is driving in the sample farmland.

[0127] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon. When executed by a processor, the computer program is implemented to perform the farmland condition identification method provided by the above methods. The method includes: acquiring static data corresponding to agricultural machinery; the static data being used to characterize the vehicle features of the agricultural machinery itself; acquiring operational status data of the agricultural machinery while it is driving in the farmland to be identified; the operational status data being time-series data with time correlation; and identifying the farmland condition to be identified based on the static data, the operational status data, and a preset neural network model, thereby determining the target farmland condition category corresponding to the farmland to be identified. The neural network model is trained based on a sample dataset, which includes multiple sample data pairs and a farmland condition category label corresponding to each sample data pair. Each sample data pair includes sample static data of the agricultural machinery and sample operational status data of the agricultural machinery while it is driving in the sample farmland.

[0128] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0129] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0130] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for identifying farmland conditions, characterized in that, include: Obtain the static data corresponding to the agricultural machinery; The static data is used to characterize the vehicle features of the agricultural machinery itself; The operation status data of the agricultural machinery while it is operating in the farmland to be identified is obtained; the operation status data is time-series data with time correlation. Based on the static data, the operational status data, and the preset neural network model, the land condition of the farmland to be identified is determined, and the target land condition category corresponding to the farmland to be identified is determined. The neural network model is trained based on a sample dataset, which includes multiple sample data pairs and the land condition category label of the sample farmland corresponding to each sample data pair. Each sample data pair includes sample static data of the agricultural machinery and sample operating status data of the agricultural machinery when it is driving in the sample farmland. The neural network model includes a single-layer perceptron and a recurrent neural network model. The step of identifying the land condition of the farmland to be identified based on the static data, the operational status data, and the preset neural network model, and determining the target land condition category corresponding to the farmland to be identified, includes: The static data is input into the single-layer perceptron for feature extraction to determine the first feature corresponding to the static data; The operational status data is input into the recurrent neural network model for feature extraction to determine the second feature corresponding to the operational status data; the second feature is time-dependent and related to the land condition of the farmland to be identified. Based on the first feature and the second feature, the land condition of the farmland to be identified is identified, and the target land condition category corresponding to the farmland to be identified is determined.

2. The farmland condition identification method according to claim 1, characterized in that, The neural network model further includes a first fully connected layer, a second fully connected layer, and an activation layer. The step of identifying the land condition of the farmland to be identified based on the first feature and the second feature, and determining the target land condition category corresponding to the farmland to be identified, includes: The first feature and the second feature are input into the first fully connected layer for feature fusion processing to determine the first fused feature; The first fused feature is input into the second fully connected layer for feature fusion processing to determine the second fused feature; The second fusion feature is input into the activation layer for classification processing to determine the target land condition category corresponding to the farmland to be identified.

3. The farmland condition identification method according to claim 1, characterized in that, The recurrent neural network model includes a long short-term memory network model, and the long short-term memory network model includes a multi-layer long short-term memory network.

4. The farmland condition identification method according to any one of claims 1 to 3, characterized in that, The step of acquiring the operating status data of the agricultural machinery while it is operating in the farmland to be identified includes: Acquire the initial operating status data of the agricultural machinery when it is driving in the farmland to be identified; the initial operating status data includes multiple initial sub-data ordered by time; Data cleaning and normalization are performed on multiple initial sub-data of the initial running state data to obtain multiple pre-processed sub-data corresponding to the initial running state data; The preprocessed sub-data of the initial running state data is filtered to determine the filtered sub-data corresponding to the initial running state data. The multiple sub-data after filtering the initial operating status data are combined to determine the operating status data of the agricultural machinery when it is driving in the farmland to be identified.

5. The farmland condition identification method according to any one of claims 1 to 3, characterized in that, The method further includes: Obtain a sample dataset; the sample dataset includes multiple sample data pairs and the land condition category label of the sample farmland corresponding to each sample data pair. Each sample data pair includes sample static data of the agricultural machinery and sample operating status data of the agricultural machinery when it is driving in the sample farmland. For each of the sample data pairs, the predicted land condition category corresponding to the sample data pair is determined based on the sample static data and sample operating status data of the agricultural machinery in the sample data pair, as well as the initial neural network model. Based on the predicted land condition category and corresponding land condition category label of the sample data pair, the initial neural network model is trained using a regularization method to determine the neural network model.

6. The farmland condition identification method according to any one of claims 1 to 3, characterized in that, The method further includes: Based on the target land condition category corresponding to the farmland to be identified, the control parameters for the agricultural machinery to travel in the farmland to be identified are dynamically adjusted; the control parameters include the agricultural machinery's travel speed and / or travel direction.

7. A farmland condition identification device, characterized in that, include: The static data acquisition module is used to acquire the static data corresponding to agricultural machinery; The static data is used to characterize the vehicle features of the agricultural machinery itself; The operational status data acquisition module is used to acquire operational status data of the agricultural machinery while it is operating in the farmland to be identified; the operational status data is time-series data with time correlation. The land condition identification module is used to identify the land condition of the farmland to be identified based on the static data, the operating status data, and a preset neural network model, and determine the target land condition category corresponding to the farmland to be identified; wherein, the neural network model is trained based on a sample dataset, the sample dataset includes multiple sample data pairs and a land condition category label of the sample farmland corresponding to each sample data pair, and each sample data pair includes sample static data of the agricultural machinery and sample operating status data of the agricultural machinery when it is driving in the sample farmland; The neural network model includes a single-layer perceptron and a recurrent neural network model. The land condition identification module is specifically used to input the static data into the single-layer perceptron for feature extraction to determine the first feature corresponding to the static data; input the operating status data into the recurrent neural network model for feature extraction to determine the second feature corresponding to the operating status data; the second feature is time-related and related to the land condition of the farmland to be identified; based on the first feature and the second feature, the land condition of the farmland to be identified is identified to determine the target land condition category corresponding to the farmland to be identified.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the farmland condition identification method as described in any one of claims 1 to 6.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the farmland condition identification method as described in any one of claims 1 to 6.

Images