Morse code intelligent recognition method and device

Abstract

The application discloses a Morse code intelligent identification method, comprising the following steps: receiving and saving a complete Morse code audio message; dividing a plurality of audio message segments according to the signal-to-noise ratio of different stages of the Morse code audio message and respectively buffering; determining a corresponding noise reduction threshold according to the average signal-to-noise ratio of each message segment; eliminating the noise of each message segment by using the noise reduction threshold to obtain a pure Morse code each message segment bit stream number; obtaining various reference time intervals by using a trained neural network; and processing each message segment bit stream by using the obtained various reference time intervals to obtain a complete message.

Description

Technical Field

[0001] This invention belongs to the field of radio technology, and particularly relates to a method and apparatus for intelligent identification of Morse code. Background Technology

[0002] Morse code is a type of intermittent signal code that uses different sequences of letters, numbers, and punctuation marks to represent different characters. Unlike modern digital communication, Morse code uses only two states: 0 and 1. Its code includes five types: short dot signals "·", long signals "—" indicating pauses between dots and dashes, medium pauses between words, and long pauses between sentences. For variable-speed Morse code recognition, the reference interval needs to be continuously updated and adjusted based on the received Morse code data to determine the dot, dash, word, and character intervals in a received Morse signal for character lookup. Currently, Morse code transcription and recognition still rely on experienced telegraph operators, which is prone to errors. Furthermore, the recognition, calculation, and decoding processes of Morse code are complex and cumbersome, resulting in low efficiency and accuracy in these areas.

[0003] Due to differences in the proficiency and usage habits of telegraph operators, the code rate, point signal, and actual duration of each signal in manually transmitted Morse code vary greatly, and the proportions are not standardized, resulting in low accuracy in Morse code recognition.

[0004] In addition, noise can be introduced during message reception due to channel transmission, equipment, and the surrounding environment, which can reduce the accuracy of Morse code recognition.

[0005] Chinese patent document CN111404846B discloses a method for receiving and identifying Morse code radio frequency signals. The method involves edge-capturing processing of the 48kHz bitstream corresponding to the received Morse code radio frequency signal to identify the rising and falling edges of the corresponding Morse code message; timing each rising and falling edge using a timer and buffering the time of each rising and falling edge; establishing a buffered high-low level duration table based on the high-level duration of dot-dash codes and the low-level duration of code intervals, character intervals, or word intervals, where the high-level duration of dot-dash codes is the time difference between the falling edge and its next adjacent rising edge, and the low-level duration of code intervals, character intervals, or word intervals is the time difference between the rising edge and its next adjacent falling edge; calculating the dot-code time length, dash-code time length, code interval, character interval, and word interval time length based on the high-low level duration table; determining the decoding time window based on the dot-code time length, dash-code time length, code interval, character interval, and word interval time length; and retrieving the ASCII code of the corresponding character of the Morse code radio frequency signal within the decoding time window using a preset Morse code ASCII lookup table. This Morse code reception and identification method is only suitable for identifying ideal, noise-free Morse code. However, in real-world wireless environments, there are numerous noise sources, and the noise is random, which significantly affects the accuracy of Morse code identification. Furthermore, this patent does not consider the differences in code rate and point signal duration in manually transmitted Morse code, which also affects the accuracy of Morse code identification.

[0006] Chinese patent document CN111724800A discloses a Morse code audio noise reduction method. The method includes: sampling and converting an audio signal into a digital signal using an ADC; calculating the signal power of the digital signal using a sliding FFT algorithm; calculating the signal-to-energy ratio (SGR) of the digital signal using the power spectrum; determining a signal threshold by calculating an estimated average SGR of the useful signal; enhancing the useful signal and suppressing noise signals based on the threshold; finding the maximum signal energy point and updating the estimated average SGR; estimating the signal energy gain by judging the relationship between the maximum signal energy point and the average energy; synthesizing the digital signal to meet the data requirements of the DAC output; and performing amplitude limiting on the output digital signal before it enters the DAC unit for conversion. This Morse code audio noise reduction method calculates the estimated average SGR of the useful signal to determine the signal threshold, and enhances the useful signal and suppresses noise signals based on the threshold. However, noise is random. During message reception, the noise level fluctuates randomly. Therefore, calculating the estimated average SGR of the useful signal and determining the signal threshold without considering noise fluctuations cannot accurately identify the Morse code. Summary of the Invention

[0007] The purpose of this invention is to provide a Morse code intelligent recognition method and apparatus for solving the above-mentioned problems in the prior art.

[0008] According to a first aspect of the present invention, a Morse code intelligent recognition method includes:

[0009] Receive and save Morse code audio messages, and calculate the signal-to-noise ratio of the Morse code audio messages in real time;

[0010] The Morse code audio message is divided into multiple Morse code audio message segments according to the real-time calculated signal-to-noise ratio.

[0011] Calculate the average signal-to-noise ratio for each Morse code audio segment;

[0012] The average signal-to-noise ratio of each Morse code audio segment is determined as the noise reduction threshold for each Morse code audio segment;

[0013] Noise in each Morse code audio segment is eliminated by using the noise reduction threshold of each Morse code audio segment, and jitter removal is performed to obtain a clean bitstream of each segment.

[0014] Each segment bitstream is fed into a pre-trained neural network to obtain various reference time intervals for each segment bitstream;

[0015] Using the obtained reference time intervals, message identification processing is performed on each pure message segment bitstream to obtain the message of each message segment. Then, each message segment message is concatenated to output the complete message.

[0016] Preferably, the neural network is a Long Short-Term Memory (LSTM) network suitable for processing time series data; the LSTM is designed as a classification model, and the output layer of the classification model has multiple nodes that output various reference time intervals such as dots, dashes, intra-character intervals, inter-character intervals, and inter-word intervals; and a trained LSTM is obtained through supervised learning using a large number of labeled Morse code bit streams.

[0017] Preferably, message identification processing for each pure segment bitstream includes:

[0018] Based on the various reference time intervals of the message segment, the dot code, dash code, and various intervals of the message segment bit stream are determined to form the dot-dash code sequence of the message segment;

[0019] The dot-and-dash encoded sequence of the message segment is matched and decoded with a preset lookup table to obtain the message segment message.

[0020] Preferably, when dividing audio message segments according to the signal-to-noise ratio (SNR), the SNR of each audio message is sampled and calculated one by one, and the position where the SNR change exceeds a preset threshold is taken as the segmentation node, so that the SNR fluctuation within each message segment is within a preset range.

[0021] Preferably, the noise reduction threshold T is positively correlated with the average signal-to-noise ratio, satisfying T=k×SNR+b, where k is the proportionality coefficient and b is the basic threshold coefficient.

[0022] According to a second aspect of the present invention, a Morse code identification device includes:

[0023] The receiving and storage module is used to receive and save Morse code audio messages and calculate the signal-to-noise ratio of the Morse code audio messages in real time.

[0024] The segmented caching module is used to divide the Morse code audio message into multiple Morse code audio message segments according to the real-time calculated signal-to-noise ratio.

[0025] The threshold determination module is used to calculate the average signal-to-noise ratio of each Morse code audio message segment; and to determine the noise reduction threshold of each Morse code audio message segment based on the average signal-to-noise ratio of each Morse code audio message segment.

[0026] The noise reduction module is used to eliminate noise from each Morse code audio message segment by using the noise reduction threshold of each Morse code audio message segment, and to perform dithering to obtain a clean bitstream of each message segment.

[0027] A pre-trained neural network for receiving each segment bit stream and outputting various reference time intervals for each segment bit stream;

[0028] The Morse code message recognition module uses various reference time intervals to perform message recognition processing on each pure message segment bit stream, obtains each message segment message, and then concatenates each message segment message to output a complete message.

[0029] Preferably, the neural network is a Long Short-Term Memory (LSTM) network for processing time series data; the LSTM is designed as a classification model, and the output layer of the classification model has multiple nodes that output various reference time intervals such as dots, dashes, intra-character intervals, inter-character intervals, and inter-word intervals; and a trained LSTM is obtained through supervised learning using a large number of labeled Morse code bit streams.

[0030] Preferably, the Morse code message recognition module includes: an encoding recognition module, used to determine the dot code, dash code and various intervals corresponding to the message segment bit stream based on the various reference time intervals, to form a dot-dash encoding sequence of the message segment bit stream; and a decoding output module, used to match the dot-dash encoding sequence with a preset Morse code encoding lookup table to complete the decoding.

[0031] Preferably, when the segmented caching module divides audio message segments according to the signal-to-noise ratio (SNR), it samples the audio message frame by frame to calculate the SNR, and takes the position where the SNR change exceeds a preset threshold as the segmentation node, and the SNR fluctuation within each message segment is within a preset range.

[0032] Preferably, the noise reduction threshold T is positively correlated with the average signal-to-noise ratio, satisfying T=k×SNR+b, where k is a proportionality coefficient and b is the basic threshold coefficient.

[0033] Compared with the prior art, the present invention has the following advantages:

[0034] Adaptive segmented noise reduction to accommodate random noise fluctuations: This invention segments the Morse code audio message according to the signal-to-noise ratio at different stages, independently determines the noise reduction threshold for each message segment and performs noise reduction, solving the problems of existing noise reduction methods that do not consider random fluctuations in noise levels and have poor noise reduction effects. It effectively filters out various noises introduced by channel transmission, equipment and environment, and improves the foundation for signal recognition in noisy environments.

[0035] Dynamically generated reference intervals to adapt to manual variable-speed transmission: Through real-time statistical analysis of high and low level duration data, reference time intervals for dot codes, dash codes, and various intervals are dynamically generated. This can adaptively match the problems of code speed changes and non-standard dot-dash duration ratios in manually transmitted Morse code, solving the problem of low recognition accuracy caused by differences in code speed and duration in existing technologies. Attached Figure Description

[0036] Figure 1 This is a flowchart illustrating the Morse code recognition method of the present invention;

[0037] Figure 2 This is a circuit diagram of the Morse code recognition device of the present invention;

[0038] Figure 3 yes Figure 2 Circuit diagram of the Morse code message module. Detailed Implementation

[0039] Figure 1The present invention illustrates a Morse code recognition method, comprising: receiving and storing a complete Morse code audio message; dividing the Morse code audio message into multiple Morse code audio message segments based on the signal-to-noise ratio (SNR) of the stored Morse code audio message at different stages, and buffering the multiple Morse code audio message segments into multiple buffers respectively; determining a noise reduction threshold for each Morse code audio message segment based on the average SNR of each Morse code audio message segment; using the noise reduction threshold for each Morse code audio message segment to eliminate noise in each Morse code audio message segment to obtain a clean Morse code audio signal for each message segment, and then performing jitter removal processing to filter out glitches and interference in the signal to obtain a clean bitstream of each message segment; feeding each message segment bitstream into a pre-trained neural network to obtain various reference time intervals for each message segment bitstream; using the obtained various reference time intervals to perform message recognition processing on each clean message segment bitstream to obtain a message for each message segment, and then concatenating each message segment message to output a complete message.

[0040] The neural network used in this invention is a Long Short-Term Memory (LSTM) network suitable for processing time series data. This LSTM is designed as a classification model, and its output layer has multiple nodes that output various reference time intervals, such as dots, dashes, intra-character intervals, inter-character intervals, and inter-word intervals. Furthermore, it is trained by supervised learning through a large number of labeled Morse code bit streams.

[0041] The present invention performs message recognition processing on each pure message segment bitstream, including: based on various reference time intervals output from the trained LSTM, determining high-level data in the high and low level duration data table as dot code or dash code, and determining low-level data as code interval, character interval, or word interval, forming a dot-dash encoding sequence of Morse code; matching the dot-dash encoding sequence with a preset Morse code encoding lookup table to obtain the corresponding characters, numbers, or punctuation marks, completing the Morse code recognition and decoding; and concatenating the decoded characters, numbers, or punctuation marks according to the message order to output the complete Morse code message content.

[0042] Specifically, the method involves dividing the stored Morse code audio message into multiple Morse code audio message segments based on the signal-to-noise ratio (SNR) of the stored Morse code audio message at different stages. This includes: sampling the stored Morse code audio message frame by frame and calculating the SNR of the audio signal in each frame; setting an SNR change threshold, and determining a segmentation node when the SNR change value of adjacent frames exceeds the SNR change threshold; and dividing the Morse code audio message into multiple consecutive Morse code audio message segments using the segmentation nodes as boundaries, with the SNR fluctuation of the audio signal within each message segment remaining within a preset range.

[0043] Specifically, based on the average signal-to-noise ratio (SNR) of each Morse code audio segment, a noise reduction threshold is determined for each Morse code audio segment. This involves: calculating the average SNR of all sampled frames within each Morse code audio segment as the average SNR of that segment; establishing a mapping relationship between the average SNR and the noise reduction threshold, where the noise reduction threshold T is positively correlated with the average SNR, satisfying the formula T = k × SNR + b, where k is a proportionality coefficient and b is a base threshold coefficient, both k and b being calibrated according to the actual wireless channel environment; and matching a corresponding noise reduction threshold to each Morse code audio segment based on this mapping relationship.

[0044] Specifically, the noise in each Morse code audio message segment is eliminated by using a noise reduction threshold for each segment. This includes: employing an adaptive filtering algorithm, using the noise reduction threshold of each Morse code audio message segment as a filtering threshold, identifying signals below the filtering threshold as noise signals and filtering them out; and performing signal enhancement processing on the filtered effective signal to amplify the amplitude of the effective signal, thereby obtaining a clean Morse code audio signal.

[0045] Figure 2 This invention discloses a Morse code recognition device, comprising: a receiving and storage module for receiving Morse code audio messages and storing complete audio messages; a segmentation and caching module electrically connected to the receiving and storage module for sampling the stored Morse code audio messages frame by frame and calculating the signal-to-noise ratio (SNR) of each frame, dividing the audio messages into multiple segments based on SNR changes, and caching each segment into a corresponding buffer; and a threshold determination module electrically connected to the segmentation and caching module for calculating the average SNR of each audio segment and determining the threshold of each segment based on the mapping relationship between the average SNR and a noise reduction threshold. The noise reduction threshold corresponding to each segment; a noise reduction processing module, electrically connected to the threshold determination module, is used to use the noise reduction threshold of each audio message segment as a filtering threshold, and to use an adaptive filtering algorithm to eliminate noise in the message segment to obtain a clean Morse code audio signal; a pre-trained neural network is used to receive the bit stream of each message segment and output various reference time intervals of each message segment bit stream; a Morse code message recognition module uses the obtained various reference time intervals to perform message recognition processing on each clean message segment bit stream to obtain the message of each message segment, and then concatenates each message segment message to output a complete message.

[0046] The neural network is a Long Short-Term Memory (LSTM) network for processing time series data. The LSTM is designed as a classification model, and the output layer of the classification model has multiple nodes that output various reference time intervals, such as dots, dashes, intra-character intervals, inter-character intervals, and inter-word intervals. The trained LSTM is obtained through supervised learning using a large number of labeled Morse code bit streams.

[0047] The segmented caching module includes a sampling unit, a signal-to-noise ratio (SNR) calculation unit, a segmentation determination unit, and a multi-buffer unit. The sampling unit is used to sample audio messages frame by frame. The SNR calculation unit is used to calculate the SNR of the sampled signal in each frame. The segmentation determination unit is used to determine the segmentation node based on the SNR change threshold. The multi-buffer unit is used to independently cache each segmented audio message.

[0048] The noise reduction processing module includes a filtering unit and a signal enhancement unit. The filtering unit is used to perform adaptive filtering with a noise reduction threshold as a threshold to filter out noise signals. The signal enhancement unit is used to amplify the amplitude of the filtered effective signal to improve signal recognition.

[0049] The device further includes a parameter calibration module, which is electrically connected to both the threshold determination module and the dynamic adjustment module. This module is used to calibrate the proportional coefficient k and the basic threshold coefficient b in the noise reduction threshold mapping relationship, as well as parameters such as the number of samples extracted and outlier removal rules in the dynamic adjustment module, based on the actual wireless channel environment, noise type, and Morse code transmission characteristics.

[0050] Figure 3 The structure of the Morse code message recognition module is shown, including: an encoding recognition module, used to determine the dot code, dash code and various intervals corresponding to high and low levels based on various reference time intervals output from the trained LSTM, forming a dot-dash code sequence; and a decoding output module, electrically connected to the encoding recognition module, used to match the dot-dash code sequence with a preset Morse code lookup table, complete the decoding and splice the complete message content.

[0051] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These embodiments are only used to explain the present invention and are not intended to limit the scope of protection of the present invention.

[0052] The Morse code recognition method provided in this embodiment is applied to the Morse code reception and recognition scenario of civilian shortwave radio communication. In this scenario, there is channel transmission noise and environmental electromagnetic noise, and the received Morse code is manually transmitted, with non-standard fluctuations in code rate and dot duration.

[0053] Step 1: Receive and save the complete Morse code audio message

[0054] The Morse code audio messages in the shortwave channel are collected by a radio receiving device and saved to the local storage unit in WAV format. The sampling rate is set to 44.1KHz and the sampling precision is 16bit.

[0055] Step 2: Segment Morse code audio messages and cache them.

[0056] The saved WAV format audio messages are sampled frame by frame, with a frame length of 10ms and a frame shift of 5ms. The signal-to-noise ratio (SNR) of each frame is calculated. The SNR change threshold is set to 15dB. When the SNR change value of adjacent frames exceeds 15dB, it is identified as a segmentation node. Using the segmentation node as the boundary, the audio message is divided into multiple consecutive audio message segments (e.g., 3 consecutive audio message segments), which are buffered in buffer 1, buffer 2, and buffer 3 respectively. The SNR fluctuation within each message segment is controlled within 5dB.

[0057] Step 3: Determine the noise reduction threshold for each message segment.

[0058] The average signal-to-noise ratio (SNR) of the three segments was calculated separately. The average SNR of the segments in buffer 1 was SNR1 = 28 dB, the average SNR of the segments in buffer 2 was SNR2 = 19 dB, and the average SNR of the segments in buffer 3 was SNR3 = 25 dB. Based on the actual shortwave channel environment calibration ratio k = 0.02 and the basic threshold coefficient b = 0.1, according to the formula T = k × SNR + b, the noise reduction thresholds were calculated as follows: Buffer 1 noise reduction threshold T1 = 0.02 × 28 + 0.1 = 0.66 V; Buffer 2 noise reduction threshold T2 = 0.02 × 19 + 0.1 = 0.48 V; Buffer 3 noise reduction threshold T3 = 0.02 × 25 + 0.1 = 0.60 V.

[0059] Step 4: Noise reduction processing to obtain a clean audio signal

[0060] The audio message segments in the three buffers are denoised using an adaptive filtering algorithm. The denoising threshold of each message segment is used as the filtering threshold to filter out noise signals below the threshold. Then, the effective signal amplitude is amplified by 2 times by a signal enhancement unit to obtain a pure Morse code audio signal without noise interference.

[0061] Step 5: Bitstream Conversion and Debouncing

[0062] A clean Morse code audio signal is input into a bitstream converter chip and converted into a 48kHz bitstream. At the same time, an RC dejitting circuit is used to dejit the bitstream, filtering out glitches with a width of less than 0.5ms in the signal to ensure signal stability.

[0063] Step 6: Feed each segment bitstream into a pre-trained neural network to obtain various reference time intervals for each segment bitstream.

[0064] Step 7: Form a dot-dash code sequence

[0065] The high-level data of the segment bitstream is compared with various reference time intervals output from the trained neural network, such as dot code and dash code reference durations. If the duration of the high-level data of the segment bitstream is between 0.18ms and 0.22ms, it is determined to be a dot code "·"; if it is between 0.54ms and 0.66ms, it is determined to be a dash code "—". The low-level data is compared with various interval reference durations. If it is between 0.18ms and 0.22ms, it is determined to be a code interval; if it is between 0.54ms and 0.66ms, it is determined to be a character interval; if it is between 1.26ms and 1.54ms, it is determined to be a word interval. Ignoring the code interval, the character interval is used as the boundary of the coding unit, and the word interval is used as the boundary of the message word group to form a continuous Morse code dot-dash coding sequence.

[0066] Step 8: Decode and output the message

[0067] The generated dot-and-dash encoding sequence is matched and queried against a pre-set internationally recognized Morse code encoding lookup table, and each encoding unit is converted into the corresponding English letters and numbers. The decoding results are then concatenated according to the character intervals and word intervals, and finally the complete Morse code message content is output.

[0068] Correspondingly, the Morse code recognition device in this embodiment consists of a receiving and storage module, a segmented buffer module, a threshold determination module, a noise reduction processing module, a bitstream conversion module, a neural network, an encoding and recognition module, a decoding and output module, and a parameter calibration module. These modules are electrically connected via a circuit board. The core control chip is an STM32F407, and a high-precision timer is integrated into the core chip. The multi-buffer unit uses three independent SRAM buffer chips, each corresponding to a different audio message segment. The parameter calibration module completes coefficient calibration via host computer software. The calibrated parameters are stored in the chip's Flash memory and can be recalibrated and modified according to the actual usage environment.

[0069] The present invention has the following technical effects:

[0070] Adaptive segmented noise reduction to accommodate random noise fluctuations: This invention segments the Morse code audio message according to the signal-to-noise ratio at different stages, independently determines the noise reduction threshold for each message segment and performs noise reduction, solving the problems of existing noise reduction methods that do not consider random fluctuations in noise levels and have poor noise reduction effects. It effectively filters out various noises introduced by channel transmission, equipment and environment, and improves the foundation for signal recognition in noisy environments.

[0071] Automated recognition and decoding improve efficiency and accuracy: It realizes a fully automated process of Morse code from audio message reception, noise reduction, signal processing to decoding output, eliminating the need for manual transcription and calculation. It completely solves the problems of high error rate and low efficiency caused by the skill level and operating habits of personnel in the manual recognition process, and greatly improves the efficiency and accuracy of Morse code recognition.

[0072] Modular design, strong scalability and practicality: The identification device adopts a modular hardware structure, with each module functioning independently yet working closely together. The parameter calibration module supports parameter adjustment according to different wireless channel environments and noise types. At the same time, the Morse code encoding lookup table supports custom encoding extensions, making it applicable to various Morse code identification scenarios such as civilian and amateur radio, with strong practicality and scalability.

[0073] The signal processing steps have been improved to enhance signal quality: a de-jitter processing step has been added after bitstream conversion to filter out glitches and interference in the signal. At the same time, signal enhancement is performed after noise reduction to improve the identification of the effective signal. This provides a high-quality digital signal for subsequent edge capture and time statistics, further ensuring the accuracy of recognition and decoding.

[0074] Although the present invention has been described in detail above, it is not limited thereto, and those skilled in the art can make various modifications based on the principles of the present invention. Therefore, all modifications made in accordance with the principles of the present invention should be understood to fall within the protection scope of the present invention.

Claims

1. A Morse code intelligent recognition method, characterized in that, include: Receive and save Morse code audio messages, and calculate the signal-to-noise ratio of the Morse code audio messages in real time; The Morse code audio message is divided into multiple Morse code audio message segments according to the real-time calculated signal-to-noise ratio. Calculate the average signal-to-noise ratio for each Morse code audio segment; The average signal-to-noise ratio of each Morse code audio segment is determined as the noise reduction threshold for each Morse code audio segment; Noise in each Morse code audio segment is eliminated by using the noise reduction threshold of each Morse code audio segment, and dithering is performed to obtain a clean bitstream of each segment. Each segment bitstream is fed into a pre-trained neural network to obtain various reference time intervals for each segment bitstream; Using the obtained reference time intervals, message identification processing is performed on each pure message segment bitstream to obtain the message of each message segment. Then, each message segment message is concatenated to output the complete message.

2. The Morse code intelligent recognition method according to claim 1, characterized in that, The neural network is a Long Short-Term Memory (LSTM) network suitable for processing time series data. The LSTM is designed as a classification model, and the output layer of the classification model has multiple nodes that output various reference time intervals, such as dots, dashes, intra-character intervals, inter-character intervals, and inter-word intervals. The trained LSTM is obtained through supervised learning using a large number of labeled Morse code bit streams.

3. The Morse code intelligent recognition method according to claim 1, wherein the message recognition processing for each pure segment bitstream using the various reference time intervals includes: Based on various reference time intervals obtained from a pre-trained neural network, the dot code, dash code, and various intervals of the message segment bit stream are determined to form the dot-dash code sequence of the message segment. The dot-and-dash encoded sequence of the message segment is matched and decoded with a preset lookup table to obtain the message segment message.

4. The Morse code intelligent recognition method according to claim 3, characterized in that, When dividing audio message segments according to the signal-to-noise ratio (SNR), the SNR of each audio message is sampled and calculated. The segmentation node is the position where the SNR change exceeds the preset threshold, so that the SNR fluctuation within each message segment is within the preset range.

5. The Morse code intelligent recognition method according to claim 1, characterized in that, The noise reduction threshold T is positively correlated with the average signal-to-noise ratio, satisfying T=k×SNR+b, where k is the proportionality coefficient and b is the basic threshold coefficient.

6. A Morse code intelligent recognition device, characterized in that, include The receiving and storage module is used to receive and save Morse code audio messages and calculate the signal-to-noise ratio of the Morse code audio messages in real time. The segmented caching module is used to divide the Morse code audio message into multiple Morse code audio message segments according to the real-time calculated signal-to-noise ratio. The threshold determination module is used to calculate the average signal-to-noise ratio of each Morse code audio message segment; and to determine the noise reduction threshold of each Morse code audio message segment based on the average signal-to-noise ratio of each Morse code audio message segment. The noise reduction module is used to eliminate noise from each Morse code audio message segment by using the noise reduction threshold of each Morse code audio message segment, and to perform dithering to obtain a clean bitstream of each message segment. A pre-trained neural network for receiving each segment bit stream and outputting various reference time intervals for each segment bit stream; The Morse code message recognition module uses various reference time intervals to perform message recognition processing on each pure message segment bit stream, obtains each message segment message, and then concatenates each message segment message to output a complete message.

7. The Morse code intelligent recognition device according to claim 6, characterized in that, The neural network is a Long Short-Term Memory (LSTM) network for processing time series data. The LSTM is designed as a classification model, and the output layer of the classification model has multiple nodes that output various reference time intervals, such as dots, dashes, intra-character intervals, inter-character intervals, and inter-word intervals. The trained LSTM is obtained through supervised learning using a large number of labeled Morse code bitstreams.

8. The Morse code intelligent recognition device according to claim 7, characterized in that, The Morse code message recognition module includes: The encoding and recognition module is used to determine the dot code, dash code and various intervals corresponding to the segment bit stream based on various reference time intervals obtained from the trained neural network, and to form a segment bit stream dot-dash encoding sequence; The decoding output module is used to match the dot-dash encoded sequence with a preset Morse code encoding lookup table to complete the decoding.

9. The Morse code intelligent recognition device according to claim 6, characterized in that, When dividing audio message segments according to signal-to-noise ratio (SNR), the segmented caching module samples the audio message frame by frame to calculate the SNR. The position where the SNR change exceeds a preset threshold is taken as the segmentation node, and the SNR fluctuation within each message segment is within a preset range.

10. The Morse code intelligent recognition device according to claim 6, characterized in that, The noise reduction threshold T is positively correlated with the average signal-to-noise ratio, satisfying T=k×SNR+b, where k is the proportionality coefficient and b is the basic threshold coefficient.

Images