Signal speech graph labeling method and system

Abstract

The invention belongs to the technical field of data annotation, and discloses a signal speech graph annotation method and system, and the method comprises the steps: 1, reading signal data, and generating a high-resolution speech graph and a synthetic speech graph; 2, inputting the generated high-resolution speech graph into a signal detection model to generate a speech graph pre-labeling result; 3, judging the correctness of a speech graph pre-labeling result by referring to the synthesized speech graph, and correcting a wrong speech graph pre-labeling result to obtain a final speech graph labeling result; and step 4, saving the speech graph data annotation file according to the final speech graph annotation result. According to the method, the generated high-resolution speech graph is input into the signal detection model to automatically generate the speech graph pre-annotation result, the manual annotation workload can be greatly reduced, a user can judge the correctness of the speech graph pre-annotation result according to the synthesized speech graph and correct the wrong speech graph pre-annotation result, and the annotation accuracy can be improved.

Description

Technical Field

[0001] The present invention relates to the field of data annotation technology, and in particular to a signal graph annotation method and system. Background Art

[0002] In the field of radio signal reception and detection, speech graphs are used to visually display the characteristics of signals. Through speech graphs, it can be determined whether there is a signal in the graph, what type of signal it is, and the location area where the signal appears. It is widely used in the actual signal target detection engineering.

[0003] With the rapid development of artificial intelligence (AI) technology in recent years, it has been widely applied in various fields, particularly in object detection, image recognition, and image classification. Labeled image data provides a training reference for machine learning and is crucial for training neural network models. Existing image object labeling systems typically focus on single images, resulting in low labeling efficiency and time-consuming and labor-intensive labeling. Summary of the Invention

[0004] To address the challenges of large amounts of signal speech graph training data and time-consuming and labor-intensive labeling, the present invention provides a signal speech graph annotation method and system. This method utilizes a signal detection model to generate speech graph pre-annotation results, then uses the synthesized speech graph to determine the accuracy of the pre-annotation results, correcting any errors. Furthermore, the system can annotate multiple speech graphs simultaneously, significantly increasing data annotation speed and improving work efficiency. The resulting annotated data can be used for signal target detection and training and tuning the signal detection model (a neural network model).

[0005] In order to achieve the above object, the present invention adopts the following technical solutions:

[0006] In one aspect, the present invention provides a signal diagram marking method, comprising:

[0007] Step 1: Read signal data and generate high-resolution speech graphs and synthetic speech graphs;

[0008] Step 2: Input the generated high-resolution speech graph into the signal detection model to generate speech graph pre-labeling results;

[0009] Step 3: Determine the correctness of the speech graph pre-annotation results by referring to the synthetic speech graph, correct the incorrect speech graph pre-annotation results, and obtain the final speech graph annotation results;

[0010] Step 4: Save the speech-image data annotation file based on the final speech-image annotation result.

[0011] Furthermore, the step 1 includes:

[0012] Step 1.1: Read the speech-graph data files to be annotated, and read multiple high-resolution speech-graph data files and multiple synthetic speech-graph data files each time, and the number of high-resolution speech-graph data files and synthetic speech-graph data files read each time is the same;

[0013] Step 1.2: Obtain multiple floating-point data from each high-resolution speech image data file to form a one-dimensional list, and convert it into a two-dimensional matrix through matrix operations. Then, display the two-dimensional matrix data in the form of a picture to obtain a high-resolution speech image;

[0014] Step 1.3: Construct complex data based on the composition of the synthetic speech and image data file, perform fast Fourier transform on the constructed complex data to obtain the spectral energy distribution of the signal, derive the synthetic speech and image based on the spectral energy distribution, and display it.

[0015] Furthermore, in step 1.3, complex data is constructed in the following manner:

[0016] X=I+jQ

[0017] Where X is complex data, I and Q are two channels of data in the synthetic speech and image data file, and j is a complex unit.

[0018] Furthermore, the high-resolution speech graph and synthetic speech Figure 1 One to one correspondence.

[0019] Furthermore, the speech image pre-labeling result includes the signal area, the signal type in the area and the judgment probability of belonging to the signal type.

[0020] Furthermore, the step 4 includes:

[0021] According to the final speech image annotation results, the labels and speech image data are stored as dat files, and multiple groups of labels and speech image data are saved each time; each group contains 1 group of label data and 1 speech image data; all data are arranged in one dimension, and the append storage mode is used each time.

[0022] Another aspect of the present invention provides a signal diagram annotation system, comprising:

[0023] Data reading module, used to read signal data and generate high-resolution speech graphs and synthetic speech graphs;

[0024] A speech-graph pre-annotation module is used to input the generated high-resolution speech-graph into a signal detection model to generate a speech-graph pre-annotation result;

[0025] The pre-annotation result correction module is used to judge the correctness of the speech graph pre-annotation result by referring to the synthetic speech graph, and to correct the wrong speech graph pre-annotation result to obtain the final speech graph annotation result;

[0026] The label data storage module is used to save the speech-image data annotation file according to the final speech-image annotation result.

[0027] Compared with the prior art, the present invention has the following beneficial effects:

[0028] The present invention can simultaneously display high-resolution speech graphs and synthetic speech graphs in a one-to-one correspondence. Users can annotate the high-resolution speech graphs by referring to the signal results of the synthetic speech graphs. Specifically, the present invention automatically generates speech graph pre-annotation results by inputting the generated high-resolution speech graphs into the signal detection model. Users can then use the synthetic speech graphs to determine the correctness of the pre-annotation results and correct any errors, thereby improving annotation accuracy and significantly reducing the workload of manual annotation. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 This is a basic flow chart of a signal diagram marking method according to an embodiment of the present invention;

[0030] Figure 2 A schematic diagram of a data storage structure provided by an embodiment of the present invention;

[0031] Figure 3 A schematic diagram of a tag data structure provided in an embodiment of the present invention;

[0032] Figure 4 This is a schematic diagram of a signal language diagram marking system architecture according to an embodiment of the present invention;

[0033] Figure 5 This is the signal language icon annotation system interface provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0034] The present invention will be further explained below with reference to the accompanying drawings and specific embodiments:

[0035] like Figure 1 As shown, a signal diagram marking method includes:

[0036] S101: Read signal data and generate high-resolution speech graphs and synthetic speech graphs;

[0037] S102: Inputting the generated high-resolution speech graph into a signal detection model (a neural network model may be used, specifically a YOLOv5s neural network model) to generate a speech graph pre-labeling result;

[0038] S103: judging the correctness of the speech graph pre-annotation result by referring to the synthetic speech graph, correcting the incorrect speech graph pre-annotation result, and obtaining the final speech graph annotation result;

[0039] S104: Saving the speech-graph data annotation file according to the final speech-graph annotation result.

[0040] Furthermore, the S101 includes:

[0041] Step S101.1: Reading speech-graph data files to be annotated, each time reading a plurality of high-resolution speech-graph data files and a plurality of synthetic speech-graph data files, and the number of high-resolution speech-graph data files and synthetic speech-graph data files read each time being the same;

[0042] Step S101.2: Obtain multiple floating-point data from each high-resolution speech image data file to form a one-dimensional list, convert it into a two-dimensional matrix through matrix operations, and then display the two-dimensional matrix data in the form of a picture to obtain a high-resolution speech image;

[0043] Step S101.3: construct complex data based on the composition of the synthetic speech graph data file, perform fast Fourier transform on the constructed complex data to obtain the spectrum energy distribution of the signal, derive the synthetic speech graph based on the spectrum energy distribution, and display it.

[0044] As a specific implementation method, in S101, 10 high-resolution speech graph data files and 10 synthetic speech graph data files are read each time; each high-resolution speech graph data file can obtain 10,000 floating-point data to form a one-dimensional list, which is converted into a two-dimensional matrix of 80*125 through matrix operation, and then the two-dimensional matrix data is displayed in the form of a picture on the system interface using a drawing command; each synthetic speech graph data file is I and Q two-way data, and the I and Q two-way data are used to construct complex data X according to formula 1, and the constructed complex data X is fast Fourier transformed to obtain the spectrum energy distribution of the signal, and a synthetic speech graph is drawn according to the spectrum energy distribution and displayed on the system interface; the interface simultaneously displays 10 high-resolution speech graphs and 10 synthetic speech graphs (specifically, direction finding speech graphs), and the high-resolution speech graphs and synthetic speech graphs Figure 1 One to one correspondence.

[0045] Formula 1: X=I+jQ

[0046] Where: j is a complex unit.

[0047] Furthermore, the S102 includes: according to the signal speech graph generated in S101, each high-resolution speech graph data is sequentially fed into the signal detection model to obtain a pre-labeling result and displayed on the interface, wherein the pre-labeling result information includes the signal area, the signal type in the area and the judgment probability of belonging to the signal type.

[0048] Furthermore, S103 includes: the user judges whether the pre-labeling result is correct. If the pre-labeling result is correct, the user can directly adopt it. If the pre-labeling result is incorrect, the user can modify the signal area and signal type, that is, the signal area and the signal type within the area, to obtain the final speech image labeling result.

[0049] Furthermore, the step S104 includes: storing the labels and speech image data as a dat file according to the final speech image annotation result. The stored dat file can be further used for signal target detection or signal detection model training and optimization. The storage format is as follows: Figure 2 As shown, 10 sets of label and speech image data are saved each time, totaling 100180 numbers; each set of label and speech image data contains 1 set of label data and 1 speech image, totaling 10018 numbers (T1); each set of label data contains 18 numbers, which can represent 3 labeled targets, and each target is represented by 6 numbers, as shown in Figure 3 As shown, B1 represents the signal type, B1-B5 represent the area where the signal appears, and B6 is the confidence level of the automatic annotation information; each picture contains 10,000 numbers; all data are arranged in one dimension, and each save adopts the append storage mode.

[0050] The present invention automatically generates speech-graph pre-annotation results using the generated high-resolution speech-graph input signal detection model. Users can refer to the synthesized speech-graph to judge the correctness of the speech-graph pre-annotation results and correct erroneous speech-graph pre-annotation results, thereby improving the accuracy of annotation and significantly reducing the workload of manual annotation.

[0051] Based on the above embodiments, Figure 4 As shown, the present invention also proposes a signal diagram annotation system, comprising:

[0052] Data reading module, used to read signal data and generate high-resolution speech graphs and synthetic speech graphs;

[0053] A speech-graph pre-annotation module is used to input the generated high-resolution speech-graph into a signal detection model to generate a speech-graph pre-annotation result;

[0054] The pre-annotation result correction module is used to judge the correctness of the speech graph pre-annotation result by referring to the synthetic speech graph, and to correct the wrong speech graph pre-annotation result to obtain the final speech graph annotation result;

[0055] The label data storage module is used to save the speech-image data annotation file according to the final speech-image annotation result.

[0056] The specific functions in each module correspond to the functions of each step in a signal diagram notation method and will not be repeated here.

[0057] As an implementation method, the signal language diagram marking system interface of the present invention is as follows: Figure 5 shown.

[0058] According to the original signal data file generated by the business system, the signal speech graph data annotation system of the present invention is used to perform data annotation. First, the original signal data file is read, and 10 high-resolution speech graph (signal) data files and 10 synthetic speech graph (signal) data files are read each time. The synthetic speech graph is used as a reference graph for signal annotation and is compared with the high-resolution speech graph. Figure 1 One-to-one correspondence; the system includes an initial detection model that can generate speech-graph pre-labeling results. Users judge the correctness of the pre-detection results based on the displayed speech-graph and modify the pre-labeling error information. After completion, the system saves the labels and speech-graph data. It can complete the labeling of 10 speech-graphs at one time. Combined with the reference of the synthetic speech-graph, it ensures both the efficiency of labeling execution and the accuracy of the labeling information.

[0059] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A signal diagram marking method, characterized in that: include: Step 1: Read signal data and generate high-resolution speech graphs and synthetic speech graphs; Step 2: Input the generated high-resolution speech graph into the signal detection model to generate speech graph pre-labeling results; Step 3: Determine the correctness of the speech graph pre-annotation results by referring to the synthetic speech graph, correct the incorrect speech graph pre-annotation results, and obtain the final speech graph annotation results; Step 4: Save the speech-image data annotation file based on the final speech-image annotation result.

2. A signal diagram marking method according to claim 1, characterized in that: The step 1 comprises: Step 1.1: Read the speech-graph data files to be annotated, and read multiple high-resolution speech-graph data files and multiple synthetic speech-graph data files each time, and the number of high-resolution speech-graph data files and synthetic speech-graph data files read each time is the same; Step 1.2: Obtain multiple floating-point data from each high-resolution speech image data file to form a one-dimensional list, and convert it into a two-dimensional matrix through matrix operations. Then, display the two-dimensional matrix data in the form of a picture to obtain a high-resolution speech image; Step 1.3: Construct complex data based on the composition of the synthetic speech and image data file, perform fast Fourier transform on the constructed complex data to obtain the spectral energy distribution of the signal, derive the synthetic speech and image based on the spectral energy distribution, and display it.

3. A signal diagram marking method according to claim 2, characterized in that: In step 1.3, the complex data is constructed in the following manner: X=I+jQ Where X is complex data, I and Q are two channels of data in the synthetic speech and image data file, and j is a complex unit.

4. A signal diagram marking method according to claim 2, characterized in that: The high-resolution speech graph and the synthetic speech graph correspond one to one.

5. A signal diagram marking method according to claim 1, characterized in that: The speech image pre-labeling result includes the signal area, the signal type in the area and the judgment probability of belonging to the signal type.

6. A signal diagram marking method according to claim 5, characterized in that: The step 4 comprises: According to the final speech image annotation results, the labels and speech image data are stored as dat files, and multiple groups of labels and speech image data are saved each time; each group contains 1 group of label data and 1 speech image data; all data are arranged in one dimension, and the append storage mode is used each time.

7. A signal diagram marking system, characterized in that: include: Data reading module, used to read signal data and generate high-resolution speech graphs and synthetic speech graphs; A speech-graph pre-annotation module is used to input the generated high-resolution speech-graph into a signal detection model to generate a speech-graph pre-annotation result; The pre-annotation result correction module is used to judge the correctness of the speech graph pre-annotation result by referring to the synthetic speech graph, and to correct the wrong speech graph pre-annotation result to obtain the final speech graph annotation result; The label data storage module is used to save the speech-image data annotation file according to the final speech-image annotation result.

Images