KR102173241B1 - Method and system for detecting abnormal sign in construction site - Google Patents

Abstract

The present invention relates to a construction site abnormal symptom detection method and system, wherein the construction site abnormal symptom detection system according to the present invention comprises: an acoustic analyzer for receiving an acoustic signal from a construction site and transmitting the acoustic signal; And an acoustic analysis server for receiving the acoustic signal of the construction site and detecting an abnormal symptom by comparing the received acoustic signal with normal sound or accident sound data, wherein the acoustic analysis server includes the characteristic of the acoustic signal. Is extracted and classified by category to construct a learning algorithm for normal and abnormal signals.

Description

Construction site abnormal symptoms detection method and system {METHOD AND SYSTEM FOR DETECTING ABNORMAL SIGN IN CONSTRUCTION SITE}

An embodiment of the present invention relates to a method and system for detecting abnormal signs in a construction site.

In the construction industry, where the number of deaths from industrial accidents reaches 500 every year, the mortality rate per 10,000 workers is 1.47, which is about 45% higher than the average mortality rate from industrial accidents of 1.01 (Korea Safety and Health Agency 2015).

Construction, excavation, and reinforced concrete construction were the main types of accidents by construction type, and as a result of investigation of the accidents of collapse and collapse, most of them collapsed during concrete pouring or due to upper load. That is, many accidents occur in temporary construction, excavation, and reinforced concrete construction, and collapse and collapse accidents occurred due to the inability of the formwork and the movable to withstand the upper load during concrete pouring.

Signs of safety accidents like this can be identified through sound, and as in Heinrich's Law, a major accident does not occur accidentally or suddenly at any moment. There are several warning signs and signs over a period of time before a major accident occurs.

In addition, even if an accident has occurred, a quick response is possible if it can be detected through sound.

Sound is a wave in which the vibration of an object is transmitted by a medium, and it is divided into Voice and Acoustics, which are human voices or speech.

Existing studies using speech analysis are mainly conducted in the direction of educational use by analyzing characteristics such as rhyme change and intonation speech speed in humanities (Gyucheol Yoon et al. 2010, Seobae Lee 2014), and in medical engineering, laryngeal cancer detection and voice treatment It is being progressed with the aim of developing a medical system for the back (Kim Kyung-Tae et al. 2002).

In the field of information and communication, artificial intelligence such as Siri and a technology to match a character's facial expressions and voices in animation games through voice recognition are being developed (Jehong Yoon et al. 2006). It is used not only to detect the situation around automobiles (Junhee Cho et al. 1997), but also to identify symptoms of diseases and abnormalities through the cry of livestock (Jongwook Lee et al. 2014), and to determine whether the same person is the same person for scientific investigation in criminal cases (Hyeonyeol Jeong 2001) Or, voice analysis technology is being applied to respond to disaster situations by judging people's emotions or situations (Lee Gyu-hwan et al., 2016).

The traffic accident automatic detection system (Hyung-Seok Lee et al. 2004, Korea Institute of Construction Technology 2016) installs CCTV in a designated area, analyzes data in real time, automatically detects accidents in tunnels, and provides information to prevent secondary accidents. Is to do.

However, since this study was applied to a tunnel, which is a linear space, it cannot be applied to a construction site in a three-dimensional space. The landslide monitoring system in the civil engineering field (Kook-Jin Kim, 2015) detects water temperature changes with a temperature measuring instrument and predicts the risk in advance by analyzing the sound when the ground is pushed. Not only does it require drilling in the ground, but also blocks external noise after drilling. It is difficult to apply due to the nature of construction sites where various sounds are generated outdoors because it detects the sound of an accident in the state of being damaged.

In addition, as a conventional technique, there is a study (Cheng at al. 2017) that tracks and monitors the work condition and productivity of heavy equipment through sound analysis, but the heavy equipment monitoring analyzes only a single machine activity in a state slightly away from external noise. The overall acoustic analysis was not performed, and there was a great limitation in applying it to the safety management field of the construction site as there was no consideration of the setup and layout (position, distance, direction, number of microphones) of the device for measurement.

The present invention was conceived to solve the above-described problem, and according to the present invention, it is intended to be able to quickly and accurately identify safety accidents in the field by comparing normal sounds and abnormal sounds to determine abnormal symptoms.

In addition, according to the present invention, before an accident occurs in a construction site, it is intended to prevent safety accidents in advance or to enable immediate response by identifying abnormal sound signs such as collapse signs and identifying the location where the sound is emitted.

A construction site abnormal symptom detection system according to the present embodiment for solving the above-described problem includes: an acoustic analyzer for receiving an acoustic signal from a construction site and transmitting the acoustic signal; And an acoustic analysis server for receiving the acoustic signal of the construction site and comparing the received acoustic signal with normal sound or accident sound data to detect an abnormality, wherein the acoustic analysis server extracts features of the acoustic signal An acoustic signal learning unit configured to a learning algorithm for normal sounds and abnormal signals by classifying them by category; An anomaly symptom analysis unit that extracts the characteristics of the acoustic signal and determines an abnormal symptom by analyzing the characteristics of the acoustic signal using a classification model constructed using a learning algorithm for the normal sound and the abnormal signal; And a position analysis unit for measuring and analyzing the occurrence position of the abnormal symptom based on the position of the sound analyzer, wherein the sound signal learning unit determines the effectiveness of a training set using the sound signal to learn a model. Learning using a validation set used for testing and a test set testing the learned model, the acoustic signal learning unit or the anomaly symptom analysis unit MFCC (Mel frequency cepstral coefficients), chroma STFT Extract features of the acoustic signal using (chromagram from a waveform or power spectrogram), melspectrogram (Mel-scaled power spectrogram), spectral contrast or tonnetz (tonal centroid features), or STFT, Spectrogram, Mel-filtering, and Log power spectrogram The first feature may be extracted using and the second feature may be extracted using the convolution layer to extract the features of the acoustic signal.
According to another embodiment of the present invention, the acoustic analyzer includes an acoustic signal input unit for collecting the amplitude, frequency, waveform or pattern of the acoustic signal of the construction site; A memory unit for storing the sound signal; An acoustic signal transmission unit for transmitting the stored acoustic signal; A location providing unit that provides an input location of an acoustic signal using location information of the acoustic analyzer and a signal arrival time; And a danger display unit for displaying a danger upon receiving an abnormal symptom detection signal from the acoustic analysis server.

delete

delete

delete

According to another embodiment of the present invention, the acoustic signal learning unit or the abnormal symptom analysis unit may pre-process the acoustic signal into a waveplot, a spectrogram, or a log power spectrogram. have.

According to another embodiment of the present invention, the position analysis unit may measure and analyze the occurrence position of the abnormal symptom by applying a position estimation algorithm to the installation position coordinate of the acoustic analyzer and the signal arrival time.

delete

delete

According to another embodiment of the present invention, the abnormal symptom analysis unit generates point history data based on the input position of the sound signal, and analyzes the sound signal by the sound analyzer, and another sound analyzer adjacent to the sound analyzer. Section history pattern data is generated by using the information obtained by analyzing the acoustic signal detected in, and abnormal signs of a corresponding construction site may be determined using the section history pattern data.

A method for detecting abnormal signs of a construction site according to an embodiment of the present invention includes: a first step of receiving, by an acoustic analyzer, an acoustic signal of a construction site and transmitting the acoustic signal; A second step of the acoustic analysis server receiving the acoustic signal of the construction site; A third step of extracting, by the acoustic analysis server, features of the acoustic signal; A fourth step of analyzing, by the acoustic analysis server, characteristics of the acoustic signal extracted using a classification model constructed using a learning algorithm for normal and abnormal signals; A fifth step of measuring and analyzing, by the acoustic analysis server, a location of a danger signal by applying a location estimation algorithm to the installation location coordinates of the acoustic analyzer and the signal arrival time; A sixth step of transmitting an abnormal symptom detection signal to the acoustic analyzer when the acoustic signal analyzed by the acoustic analysis server corresponds to the danger signal; And a seventh step in which the acoustic analyzer receives the abnormal symptom detection signal and displays a danger, and before the first step or after the seventh step, the acoustic analysis server extracts the characteristics of the sound signal Comprising a learning algorithm for the normal sound and the abnormal signal by dividing by category; and the third step is a training set used to train the model using the acoustic signal, and used to check the validity. Learning using a validation set and a test set that tests the trained model, MFCC (Mel frequency cepstral coefficients), chroma STFT (chromagram from a waveform or power spectrogram), melspectrogram (Mel-scaled) power spectrogram), spectral contrast, or tonnetz (tonal centroid features) to extract the features of the acoustic signal, or to extract the first feature using STFT, Spectrogram, Mel-filtering, and Log power spectrogram, and the second order using a convolution layer. By extracting features, features of the acoustic signal may be extracted.
According to another embodiment of the present invention, the third step may include preprocessing, by the acoustic analysis server, the acoustic signal into a waveplot, a spectrogram, or a log power spectrogram; It may further include.

delete

delete

delete

According to an embodiment of the present invention, it is possible to quickly and accurately identify safety accidents in the field by comparing normal sounds and abnormal sounds to determine abnormal symptoms.

In addition, according to an embodiment of the present invention, before an accident occurs in a construction site, it is possible to prevent safety accidents in advance or to immediately respond by identifying abnormal acoustic signs such as collapse signs and identifying the location where the sound is emitted.

Furthermore, according to an embodiment of the present invention, the role of a manager can be expected by monitoring various situations in the field in real time.

1 is a block diagram of a construction site abnormal sign detection system according to an embodiment of the present invention.
2 is a view showing a waveform of an impact sound according to an embodiment of the present invention.
3 is a view for explaining the work noise of the construction site.
4 is a view showing a waveform of pre-processing an acoustic signal of a construction site abnormal symptom detection system according to an embodiment of the present invention.
5 is a view for explaining an acoustic signal learning method and an abnormal symptom analysis method of a construction site abnormal symptom detection system according to an embodiment of the present invention.
6 is a flowchart illustrating a method of detecting abnormal signs at a construction site according to an embodiment of the present invention.

Hereinafter, a preferred embodiment of the present invention will be described in detail with reference to the accompanying drawings. However, in describing the embodiments, when it is determined that a detailed description of a related known function or configuration may unnecessarily obscure the subject matter of the present invention, a detailed description thereof will be omitted. In addition, the size of each component in the drawings may be exaggerated for the sake of explanation, and does not mean a size that is actually applied.

1 is a block diagram of a construction site abnormal symptom detection system according to an embodiment of the present invention, and FIG. 2 is a view showing a waveform of an impact sound according to an embodiment of the present invention.

3 is a view for explaining the working noise of the construction site, Figure 4 is a view showing a waveform pre-processed the acoustic signal of the construction site abnormal symptoms detection system according to an embodiment of the present invention, Figure 5 is the present invention A diagram for explaining a method of learning an acoustic signal and a method of analyzing an abnormality in a construction site abnormality detection system according to an embodiment of

Hereinafter, a configuration of a construction site abnormal symptom detection system according to an embodiment of the present invention will be described with reference to FIGS. 1 to 5.

Referring to FIG. 1, a construction site abnormal symptom detection system according to an embodiment of the present invention may include an acoustic analyzer 110 and an acoustic analysis server 120.

The acoustic analyzer 110 receives an acoustic signal of a construction site and transmits the acoustic signal.

For example, during concrete pouring, the formwork or movement may collapse due to the lateral pressure of the concrete, resulting in personal or material damage. In addition, even if the temporary facility does not collapse, a gap may be opened by the weight and a concrete leakage accident may occur. Accidents that occur during concrete pouring need to be dealt with quickly, but it is difficult to deal with them immediately. In particular, the problem of concrete leakage accidents is that when concrete is poured on the upper floors, the accidents occurring on the lower floors are not recognized at all.

According to an embodiment of the present invention, such problems can be solved through analysis of an acoustic signal.

That is, as illustrated in FIG. 2, problem occurrence situations may be detected through analysis of the amplitude, frequency, waveform, or pattern of sound generated during concrete pouring.

FIG. 2 (a) shows a signal that generates an impact sound and a burst sound in a general construction site noise situation, and FIG. 2 (b) shows a signal that generates an impact sound and a burst sound during pouring.

In (a) of FIG. 2, it can be seen that the impact sound and the bursting sound are clearly distinguished. This is because the ambient noise is small, so that the impact sound and the bursting sound having a high decibel level generally appear clearly.

2B is a pouring situation, and a very loud noise is generated during the pouring, and as a result of frequency analysis, it can be seen that a unique pattern and size can be shown and distinguished even in a noise situation in which the impact sound and the burst sound are large. At this time, it can also be confirmed that the impact sound and the burst sound have different threshold values from the general noise.

In addition, Figure 3 is for explaining the working noise of the construction site, Figure 3 (a) is a normal noise level fluctuation is small and constant, Figure 3 (b) is a fluctuating noise, the level is irregular and continuously It changes in a certain range, and FIG. 3(c) is a continuous impact noise, which has an extremely short duration, and FIG. 3(d) is a repetitive impact noise, which has a very short duration, and FIG. 3(e) ) Is an intermittent noise, which occurs intermittently and has a duration of several seconds or more, Fig. 3(f) shows the noise generated as a separate impact noise is independently separated, and Fig. 3(g) is a quasi-normal impact noise. It has a characteristic that a certain level of noise is generated repeatedly at extremely short time intervals.

In order to analyze various acoustic signals as described above, according to an embodiment of the present invention, the acoustic analyzer 110 may be installed under the concrete placement layer to obtain an acoustic signal, and the acoustic analyzer 110 It may be configured to be portable as possible, so that it can be easily moved by moving the pouring area or moving the floor.

More specifically, the acoustic analyzer 110 according to an embodiment of the present invention may be configured to include an acoustic signal input unit 111, a memory unit 112, an acoustic signal transmission unit 113, and a danger display unit 115. have.

The sound signal input unit 111 collects the amplitude, frequency, waveform or pattern of the sound signal of the construction site, the memory unit 112 stores the sound signal, and the sound signal transmission unit 113 is the stored sound signal Can be transmitted.

In addition, the danger display unit 115 may be configured with an LED indicator to indicate danger when receiving an abnormal symptom detection signal from the acoustic analysis server 120.

The acoustic analysis server 120 receives the acoustic signal of the construction site, compares the received acoustic signal with normal sound or accident sound data to detect abnormal signs, and more specifically, the acoustic analysis server 120 It may be configured to include a signal learning unit 121, an abnormal symptom analysis unit 122, and a location analysis unit 123.

The acoustic signal learning unit 121 may extract features of the acoustic signal and divide it into categories to construct a learning algorithm for normal and abnormal signals.

Acoustic signals can be different from mechanical equipment used in construction sites, and in order to prevent learning into a model that performs well only in certain situations due to the limitation of the data set, the amount of data belonging to each classification category is balanced. To collect data.

In addition, the acoustic signal learning unit 121 is a training set used to train the acoustic signal model, a validation set used to check the validity, and a test set for testing the trained model. Set) can be used to learn.

In this way, for learning of an acoustic signal, the acoustic signal learning unit 121 may pre-process the acoustic signal into an image, and more specifically, the acoustic signal may be converted to a waveplot, a spectrogram, or a logarithmic signal. It can be preprocessed with a log power spectrogram.

FIG. 4 is a view showing a waveform obtained by preprocessing an acoustic signal of a construction site abnormal symptom detection system according to an embodiment of the present invention, and FIG. 4A is a view obtained by preprocessing the acoustic signal with a waveplot, FIG. 3(b) is a diagram preprocessed with a spectrogram, and FIG. 4(c) is a diagram preprocessed with a log power spectrogram.

The acoustic signal learning unit 121 may extract features from data preprocessed into an image as described above.

Referring to FIG. 5, the acoustic signal learning unit 121 constructs a learning algorithm 126 for normal and abnormal signals by extracting features from an input acoustic signal using a feature extractor 125 when learning an acoustic signal. do.

More specifically, the feature extractor 125 uses MFCC (Mel frequency cepstral coefficients), chroma STFT (chromagram from a waveform or power spectrogram), melspectrogram (Mel-scaled power spectrogram), spectral contrast, or tonnetz (tonal centroid features). It is possible to extract features of the acoustic signal.

Alternatively, the feature extractor 125 is configured to extract the first feature using STFT, Spectrogram, Mel-filtering, and Log power spectrogram, and extract the second feature using a convolution layer to extract the features of the acoustic signal. I can.

That is, the acoustic signal learning unit 121 may construct a learning algorithm 126 for normal and abnormal signals by extracting features using the feature extractor 125 configured as described above.

On the other hand, the abnormal symptom analysis unit 122 may determine the abnormal symptom by extracting the characteristics of the sound signal and analyzing the characteristics of the sound signal using a classification model configured using a learning algorithm for the normal sound and the abnormal signal. Referring to FIG. 5, the anomaly symptom analysis unit 122 extracts a feature from an input sound signal using a feature extractor 125, and classifies it through a classification model 128 based on the extracted feature, Signal analysis and detection of abnormal signs are possible.

The feature extractor 127 uses MFCC (Mel frequency cepstral coefficients), chroma STFT (chromagram from a waveform or power spectrogram), melspectrogram (Mel-scaled power spectrogram), spectral contrast, or tonnetz (tonal centroid features) using the acoustic signal. Or alternatively, it is configured to extract the first feature using STFT, Spectrogram, Mel-filtering, and Log power spectrogram, and extract the second feature using the Convolution Layer to extract the features of the acoustic signal. I can.

That is, the abnormal symptom analysis unit 122 extracts a feature using the feature extractor 127 configured as described above, and uses a classification model configured using the learning algorithm 126 for the normal sound and the abnormal signal to feature the acoustic signal. Can be analyzed to determine the abnormality.

In addition, the location analysis unit 123 may measure and analyze the location of the abnormal symptom based on the location of the acoustic analyzer 110.

More specifically, the location analysis unit 123 may measure and analyze the occurrence location of the abnormal symptom by applying a location estimation algorithm to the installation location coordinate of the acoustic analyzer 110 and the signal arrival time.

6 is a flowchart illustrating a method for detecting abnormal signs in a construction site according to an embodiment of the present invention.

Hereinafter, a method for detecting abnormal signs at a construction site according to an embodiment of the present invention will be described with reference to FIG. 6.

First, an acoustic analyzer is installed at a construction site to receive an acoustic signal from the construction site, collect acoustic information, and transmit the acoustic signal (S510).

In addition, the acoustic analysis server may receive the acoustic signal of the construction site, extract the characteristics of the acoustic signal, and the acoustic analysis server extracted using a classification model configured using a learning algorithm for normal and abnormal signals. The characteristics of the sound signal may be analyzed (S520).

In addition, in order to analyze the characteristics of the acoustic signal, the acoustic analysis server may pre-process with a waveplot, a spectrogram, or a log power spectrogram.

In addition, the acoustic analysis server uses MFCC (Mel frequency cepstral coefficients), chroma STFT (chromagram from a waveform or power spectrogram), melspectrogram (Mel-scaled power spectrogram), spectral contrast, or tonnetz (tonal centroid features) using the acoustic signal. In addition, the feature extractor extracts the first feature using STFT, Spectrogram, Mel-filtering, and Log power spectrogram, and extracts the second feature using a convolution layer to extract the features of the acoustic signal. It can be configured to extract.

In addition, the acoustic analysis server may measure and analyze the location of the danger signal by applying a location estimation algorithm to the installation location coordinate of the acoustic analyzer and the signal arrival time (S530).

Thereafter, if the acoustic signal analyzed by the acoustic analysis server corresponds to a danger signal, it is determined as an abnormal situation and the location of the danger signal is provided, and the abnormal symptom detection signal is transmitted to the acoustic analyzer, and the acoustic analyzer May receive the abnormal symptom detection signal to store the abnormal situation and display the danger (S540).

In addition, according to an embodiment of the present invention, in order to analyze such abnormal symptoms, the acoustic analysis server extracts features of the acoustic signals and divides them into categories to construct a learning algorithm for normal sounds and abnormal signals. have.

That is, when the acoustic analysis server learns an acoustic signal, a feature is extracted from the input acoustic signal using a feature extractor to construct a learning algorithm for normal and abnormal signals.

More specifically, the feature extractor of the acoustic analysis server is MFCC (Mel frequency cepstral coefficients), chroma STFT (chromagram from a waveform or power spectrogram), melspectrogram (Mel-scaled power spectrogram), spectral contrast or tonnetz (tonal centroid features) The characteristics of the acoustic signal can be extracted using STFT, Spectrogram, Mel-filtering, and Log power spectrogram, and the second feature is extracted using a convolution layer. Can be configured to extract.

That is, the acoustic analysis server constructs a learning algorithm for normal and abnormal signals by extracting features using the feature extractor 125 configured as described above, and uses a learning algorithm for normal and abnormal signals configured as described above. Thus, by constructing a classification model, an abnormal symptom can be determined by analyzing characteristics of an acoustic signal.

Therefore, according to an embodiment of the present invention, it is possible to quickly and accurately identify safety accidents in the site by determining abnormal symptoms by comparing normal sounds and abnormal sounds, and abnormal acoustic signs such as collapse signs before an accident occurs in a construction site. It is possible to prevent safety accidents in advance or to respond immediately by identifying the location and the location where the sound is emitted.

In the detailed description of the present invention as described above, specific embodiments have been described. However, various modifications may be made without departing from the scope of the present invention. The technical idea of the present invention is limited to the above-described embodiments of the present invention and should not be defined, and should not be defined by the claims as well as the claims and their equivalents.

110: acoustic analyzer
111: sound signal input unit
112: memory unit
113: sound signal transmission unit
115: danger display
120: acoustic analysis server
121: acoustic signal learning unit
122: abnormal symptom analysis unit
123: position analysis unit
125: feature extractor
126: learning algorithm
127: feature extractor
128: classification model

Claims (12)

An acoustic analyzer for receiving an acoustic signal of a construction site and transmitting the acoustic signal; And
An acoustic analysis server for receiving the acoustic signal of the construction site and comparing the received acoustic signal with normal sound or accident sound data to detect abnormal signs;
Including,
The acoustic analysis server,
An acoustic signal learning unit that extracts the characteristics of the acoustic signal and divides it by category to construct a learning algorithm for normal and abnormal signals;
An anomaly symptom analysis unit that extracts the characteristics of the acoustic signal and determines an abnormal symptom by analyzing the characteristics of the acoustic signal using a classification model constructed using a learning algorithm for the normal sound and the abnormal signal; And
A position analysis unit measuring and analyzing the occurrence position of the abnormal symptom based on the position of the acoustic analyzer;
Including,
The acoustic signal learning unit,
The acoustic signal is trained using a training set used to train a model, a validation set used to check the validity, and a test set to test the trained model,
The acoustic signal learning unit or the abnormal symptom analysis unit,
Extracting features of the acoustic signal using MFCC (Mel frequency cepstral coefficients), chroma STFT (chromagram from a waveform or power spectrogram), melspectrogram (Mel-scaled power spectrogram), spectral contrast, or tonnetz (tonal centroid features), or
A construction site abnormal symptom detection system that extracts the first feature using STFT, Spectrogram, Mel-filtering and Log power spectrogram, and extracts the features of the acoustic signal by extracting the second feature using the Convolution Layer.
The method according to claim 1,
The acoustic analyzer,
An acoustic signal input unit for collecting the amplitude, frequency, waveform or pattern of the acoustic signal of the construction site;
A memory unit for storing the sound signal;
An acoustic signal transmission unit for transmitting the stored acoustic signal;
A danger display unit for displaying a danger upon receiving an abnormal symptom detection signal from the acoustic analysis server;
Construction site abnormality detection system comprising a further.
delete delete The method according to claim 1,
The acoustic signal learning unit or the abnormal symptom analysis unit,
A construction site abnormal symptom detection system that pre-processes the acoustic signal into a waveplot, spectrogram, or log power spectrogram.
delete delete The method according to claim 1,
The location analysis unit,
A construction site abnormal symptom detection system that measures and analyzes the occurrence position of the abnormal symptom by applying a position estimation algorithm to the installation position coordinate of the acoustic analyzer and the signal arrival time.
A first step of receiving, by an acoustic analyzer, an acoustic signal of a construction site and transmitting the acoustic signal;
A second step of the acoustic analysis server receiving the acoustic signal of the construction site;
A third step of extracting, by the acoustic analysis server, features of the acoustic signal;
A fourth step of analyzing, by the acoustic analysis server, characteristics of the acoustic signal extracted using a classification model constructed using a learning algorithm for normal and abnormal signals;
A fifth step of measuring and analyzing, by the acoustic analysis server, a location of a danger signal by applying a location estimation algorithm to the installation location coordinates of the acoustic analyzer and the signal arrival time;
A sixth step of transmitting an abnormal symptom detection signal to the acoustic analyzer when the acoustic signal analyzed by the acoustic analysis server corresponds to the danger signal; And
A seventh step of indicating, by the acoustic analyzer, a danger by receiving the abnormal symptom detection signal;
Including,
Before the first step or after the seventh step,
The acoustic analysis server extracting features of the acoustic signal and classifying it by category to construct a learning algorithm for normal and abnormal signals;
Including more,
The third step,
The acoustic signal is trained using a training set used to train a model, a validation set used to check the validity, and a test set to test the trained model,
Extracting features of the acoustic signal using MFCC (Mel frequency cepstral coefficients), chroma STFT (chromagram from a waveform or power spectrogram), melspectrogram (Mel-scaled power spectrogram), spectral contrast, or tonnetz (tonal centroid features), or
A construction site abnormal symptom detection method that extracts the first feature using STFT, Spectrogram, Mel-filtering and Log power spectrogram, and extracts the second feature using the Convolution Layer to extract the features of the acoustic signal.
The method of claim 9,
The third step,
Pre-processing, by the acoustic analysis server, the acoustic signal into a waveplot, a spectrogram, or a log power spectrogram;
Construction site abnormality detection method further comprising a.
delete delete

Images