CN111461252A

CN111461252A - Chick sex detector and detection method

Info

Publication number: CN111461252A
Application number: CN202010284618.0A
Authority: CN
Inventors: 汪文洋; 殷蔚明; 王伦; 刘松明; 徐琼; 周琼
Original assignee: China University of Geosciences
Current assignee: China University of Geosciences
Priority date: 2020-04-13
Filing date: 2020-04-13
Publication date: 2020-07-28

Abstract

The invention provides a chick sex detector and a chick sex detection method, wherein the chick sex detector comprises: the audio collector collects the chick singing, obtains a chick singing audio signal and sends the chick singing audio signal to the processor; the processor processes the acquired chick sound-calling audio signal to obtain a corresponding two-dimensional spectrogram image signal; the processor judges the gender of the chick by adopting an Alexnet network algorithm according to the spectrogram image signal to obtain a final judgment result of the gender of the chick; and displaying the final judgment result of the gender of the chick on a display screen. The invention has the beneficial effects that: the detector provided by the invention simplifies the gender detection of the chicks and saves the manual labor; the detection method combines image processing and sound processing technologies, visualizes time domain and frequency domain information in the sound of the chicks, and further analyzes voiceprint characteristics presented in a spectrogram of the chicks to realize classification and identification of the sounding of the chicks, know the gender of the chicks, improve the welfare breeding of the laying hens and improve the production efficiency.

Description

Chick sex detector and detection method

Technical Field

The invention relates to the field of chick gender detection, in particular to a chick gender detector and a chick gender detection method.

Background

With the rapid development of artificial intelligence, the audio processing capability is remarkably improved. Many algorithmic models for image video based on neural networks are applicable to a variety of scenarios. The accuracy and the real-time performance of the intelligent algorithm based on deep learning are greatly improved in the aspects of specific face recognition, voice recognition of specific people, posture recognition, expression recognition, gesture recognition and the like.

The speech characteristics (pitch frequency) of the vocal organs characteristics reflect the gender difference to some extent. The waveforms of the cocks and the hens are called and the fundamental tone frequency is analyzed, and the differences exist between the cocks and the hens. The cock fundamental tone frequency is 4.8K, and the hen gene frequency is 5.2K. In addition, there is a difference in formants and other voiceprint information between rooster and hen.

The spectrogram displays a great deal of information related to the characteristics of the sound signal, such as the change of frequency domain parameters such as formants, energy and the like along with time, and has the characteristics of a time domain waveform and a spectrogram. The spectrogram is a display image of time sequence related Fourier analysis and can reflect the transformation of a sound signal frequency spectrum along with the change of time, the abscissa of the spectrogram is time, the ordinate is frequency, and a coordinate point value is voice data energy.

Disclosure of Invention

In order to solve the above problems, the present invention provides a chick sex detector, comprising: the system comprises an audio collector, a processor, a display screen, a wireless connection device, a power supply, a loudspeaker, a body, function keys and a network server;

the body is in a hollow cylindrical shape, a cavity is formed inside the body, the display screen and the function keys are embedded on the outer surface of the body, and the audio collector, the processor, the wireless connection device, the power supply and the loudspeaker are fixedly arranged in the cavity of the body;

the power supply is electrically connected with the audio collector, the processor, the display screen, the wireless connecting device and the loudspeaker respectively to supply power to the whole detector; the audio collector, the display screen, the wireless connecting device, the loudspeaker and the function keys are respectively and electrically connected with the processor; the processor communicates with the network server through the wireless connection device;

and the processor receives the trigger signal of the function key and respectively controls the audio collector, the display screen, the wireless connecting device and the loudspeaker according to the trigger signal.

Furthermore, an audio mounting hole corresponding to the audio collector is formed in one end of the body, and the audio collector is mounted at one end inside the body through the audio mounting hole; the other end of the body is provided with a loudspeaker mounting hole corresponding to the loudspeaker, and the loudspeaker is fixedly mounted at the other end in the body through the loudspeaker mounting hole.

Furthermore, the audio collector is a microphone, and an audio collecting port faces the outer side of the body through the audio mounting hole so as to collect audio conveniently; the loudspeaker port of speaker passes through the speaker mounting hole orientation the body outside to play audio frequency.

Furthermore, the processor is a raspberry processor with a memory card, the display screen is an L12864 liquid crystal display screen, and the wireless connection device is an ESP8266 serial port WIFI module.

Further, the function key includes: the saving key, the clearing key, the recording key and the starting key are common mechanical buttons or touch buttons, are respectively electrically connected with different I/O ports of the processor through signal lines, and press the corresponding keys, and the processor acquires corresponding trigger signals so as to control other equipment to respond.

Further, the power supply is a rechargeable battery with a charging interface; the charging interface is embedded in the outer surface of the body to connect an external charging wire and charge the rechargeable battery.

Further, the chick sex detection method is applied to a chick sex detector and specifically comprises the following steps:

s101: the audio collector collects the chick singing, obtains a chick singing audio signal and sends the chick singing audio signal to the processor;

s102: the processor processes the acquired chick sound-calling audio signal to obtain a corresponding two-dimensional spectrogram image signal;

s103: the processor judges the gender of the chick by adopting an Alexnet network algorithm according to the spectrogram image signal to obtain a final judgment result of the gender of the chick;

s104: and displaying the final judgment result of the gender of the chick by a display screen for viewing.

Further, in step S102, the processor converts the chick sound audio signal into a two-dimensional spectrogram image signal by using a spectogram function in matlab.

Further, in step S103, the processor determines the gender of the chick according to the spectrogram image signal by using an Alexnet network algorithm to obtain a result of determining the gender of the chick; the method specifically comprises the following steps:

s201: the processor adopts a local classifier to compare and judge the spectrogram image signals to obtain a preliminary judgment result of the gender of the chick, and the preliminary judgment result is displayed through a display screen;

the local classifier is a trained network server classifier obtained after the network server trains the convolutional neural network; the initial local classifier is directly obtained by downloading the processor in a network server through a wireless connection device;

s202: the processor sends the preliminary judgment result and the spectrogram image signal to the network server together through the wireless connection device;

s203: the network server performs online operation by adopting a convolutional neural network according to the received spectrogram image signal to obtain a final judgment result; the convolutional neural network is a pre-trained network server classifier;

s204: the network server sends the final judgment result to the processor; the processor judges whether the preliminary judgment result is consistent with the final judgment result; if yes, go to step S205; otherwise, go to step S206;

s205: the processor stores the final judgment result locally for viewing; and to step S207;

s206: the processor stores the final judgment result in the local and updates a local classifier according to the trained network server side classifier in the network server;

s207: and (6) ending.

Further, in step S203, the convolutional neural network includes a convolutional layer, a pooling layer and a fully-connected layer, which are connected in sequence, and an activation function of the convolutional neural network adopts an RE L U function, where 5 convolutional layers, 3 pooling layers and 3 fully-connected layers.

The technical scheme provided by the invention has the beneficial effects that: the detector provided by the invention simplifies the gender detection of the chicks and saves the manual labor; the detection method combines image processing and sound processing technologies, visualizes time domain and frequency domain information in the sound of the chicks, and further analyzes voiceprint characteristics presented in a spectrogram of the chicks to realize classification and identification of the sounding of the chicks, know the gender of the chicks, improve the welfare breeding of the laying hens and improve the production efficiency.

Drawings

The invention will be further described with reference to the accompanying drawings and examples, in which:

FIG. 1 is a diagram of an apparatus of a chick sex detector according to an embodiment of the present invention;

FIG. 2 is a flow chart of a method for detecting gender of a chicken in an embodiment of the present invention;

FIG. 3 is a schematic diagram of a convolutional neural network in an embodiment of the present invention;

FIG. 4 is a diagram illustrating an activation function RE L U according to an embodiment of the present invention;

FIG. 5 is a diagram of an Alexnet network model architecture in an embodiment of the present invention;

FIG. 6 is a schematic diagram of a convolution according to an embodiment of the present invention;

FIG. 7 is a schematic illustration of pooling in an embodiment of the present invention.

Detailed Description

For a more clear understanding of the technical features, objects and effects of the present invention, embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

The embodiment of the invention provides a chick sex detection method and instrument and a chick sex detection method.

Referring to fig. 1, fig. 1 is a diagram of an apparatus of a chick sex detector according to an embodiment of the present invention; the method comprises the following steps: the system comprises an audio collector 1, a processor 2, a display screen 3, a wireless connecting device 4, a power supply 5, a loudspeaker 10, a body 11, a function key 12 and a network server;

the body 11 is in a hollow cylindrical shape, a cavity is formed inside the body, the display screen 3 and the function keys 12 are embedded on the outer surface of the body 11, and the audio collector 1, the processor 2, the wireless connection device 4, the power supply 5 and the loudspeaker 10 are fixedly arranged in the cavity of the body 11;

the power supply 5 is electrically connected with the audio collector 1, the processor 2, the display screen 3, the wireless connecting device 4 and the loudspeaker 10 respectively to supply power to the whole detector; the audio collector 1, the display screen 3, the wireless connection device 4, the loudspeaker 10 and the function key 12 are respectively electrically connected with the processor 2; the processor 2 communicates with the network server through the wireless connection device 4;

the processor 2 receives the trigger signal of the function key 12 and controls the audio collector 1, the display screen 3, the wireless connection device 4 and the loudspeaker 10 respectively according to the trigger signal.

An audio mounting hole corresponding to the audio collector 1 is formed in one end of the body 11, and the audio collector 1 is mounted at one end inside the body 11 through the audio mounting hole; the other end of the body 11 is provided with a speaker mounting hole corresponding to the speaker 10, and the speaker 10 is fixedly mounted at the other end in the body 11 through the speaker mounting hole 10.

The audio collector 1 is a microphone, and an audio collecting port faces the outer side of the body 11 through the audio mounting hole so as to collect audio conveniently; the speaker port of the speaker 10 faces the outside of the body 11 through the speaker mounting hole so as to play audio.

The processor 2 is a raspberry processor with a memory card, the display screen 3 is an L12864 liquid crystal display screen, and the wireless connection device 4 is an ESP8266 serial port WIFI module.

The function key 12 includes: the save key 6, the clear key 7, the record key 8 and the power key 9 (for power on and power off) are all common mechanical buttons or touch buttons and are electrically connected with different I/O ports of the processor 2 through signal lines respectively.

When the corresponding key is pressed, the processor 2 will obtain the corresponding trigger signal, so as to control other devices to respond.

The power supply 5 is a rechargeable battery with a charging interface; the charging interface is embedded in the outer surface of the body 11 to connect an external charging wire and charge the rechargeable battery.

The charging interface is a TYPE-C charging interface (a chargeable and dischargeable lithium battery is adopted to provide 5v power supply for the whole system).

A chick sex detection method, which is applied to the chick sex detector; referring to fig. 2, fig. 2 is a flowchart of a method for detecting a gender of a chick according to an embodiment of the present invention, which specifically includes the following steps:

s101: the audio collector 1 collects the chick singing, obtains a chick singing audio signal and sends the chick singing audio signal to the processor 2;

s102: the processor 2 processes the acquired chick squeaking audio signal to obtain a corresponding two-dimensional spectrogram image signal;

s103: the processor 2 judges the gender of the chick by adopting an Alexnet network algorithm according to the spectrogram image signal to obtain a final judgment result of the gender of the chick;

s104: display screen 3 is right the final judged result of chick sex is shown for look over to report final judged result through speaker 10.

S101, an audio collector 1 collects the chick singing, obtains a chick singing audio signal and sends the chick singing audio signal to a processor 2; the method specifically comprises the following steps:

the recording key 8 is pressed, the microphone is aligned to the position near the mouth of the chick, the chick is called in a relatively quiet (the ambient noise decibel is smaller than or equal to a preset threshold) situation for about 3-5 seconds, and the microphone converts the sound signal into an electric signal and transmits the electric signal into a processor raspberry pie through two wires. The processor eliminates noise irrelevant to the singing through smooth filtering, and samples and quantizes the sound signal to obtain the singing audio signal of the chick.

In step S102, the processor 2 converts the chick squeaking audio signal into a two-dimensional spectrogram image signal by using a spectogram function in matlab.

The patterns in the spectrogram comprise transverse lines, random lines, vertical strips and the like, the transverse lines are black bands parallel to the time axis and are formants, the corresponding formant frequency and bandwidth can be determined according to the frequency and bandwidth corresponding to the transverse lines, and whether the transverse lines appear in the spectrogram of a section of audio is an important mark for judging whether the spectrogram is voiced sound or not is judged; the vertical bars are narrow black bars perpendicular to the time axis, each vertical bar corresponding to a fundamental tone, the start of a stripe corresponding to the start of a voiceprint pulse, the distance between stripes representing the fundamental tone, the denser the stripes representing the higher the frequency of the fundamental tone. Chicks of different genders have differences on the spectrogram, which provides a basis for gender identification of the chicks. The detector uses the spectrogram of the singing of the chicks as a basis. Since three-dimensional information is expressed by using a two-dimensional plane, the magnitude of the energy value is expressed by color, and the darker the color, the stronger the sound energy representing the point.

In step S103, the processor 2 determines the gender of the chick by using an Alexnet network algorithm according to the spectrogram image signal to obtain a chick gender determination result; the method specifically comprises the following steps:

s201: the processor 2 adopts a local classifier to compare and judge the spectrogram image signals to obtain a preliminary judgment result of the gender of the chick, and the preliminary judgment result is displayed through a display screen 3;

the local classifier is a trained network server classifier obtained after the network server trains the convolutional neural network; the initial local classifier is directly obtained by downloading the processor 2 in a network server through the wireless connection device 4, and the subsequent local classifier is obtained by updating the processor 2 according to the convolutional neural network trained in the network server;

s202: the processor 2 sends the preliminary judgment result and the spectrogram image signal to the network server together through the wireless connection device 4;

s204: the network server sends the final judgment result to the processor 2; the processor 2 judges whether the preliminary judgment result is consistent with the final judgment result; if yes, go to step S205; otherwise, go to step S206;

s205: the processor 2 stores the final judgment result locally for viewing; go to step S207;

s206: the processor 2 stores the final judgment result in the local and updates a local classifier according to the trained network server side classifier in the network server; (the user saves and deletes the final judgment result data in the local storage card through the save key 6 and the clear key 7 according to the actual requirement);

s207: and (6) ending.

In step S203, the Convolutional neural network (CNN, as shown in fig. 3) includes a Convolutional layer, a pooling layer, and a full-link layer, which are connected in sequence, where an activation function of the Convolutional neural network employs an RE L U function, where 5 Convolutional layers, 3 pooling layers, and 3 full-link layers, and an expression of the activation function RE L U (as shown in fig. 4) is as follows:

wherein x is a middle-level signal of the input spectrogram image signal after convolution and pooling, the output is uniformly 0 when x is less than 0, and the output is x when x is greater than 0.

The Alexnet network algorithm is divided into two parts: a network server classifier and a local classifier. The network server comprises a preset spectrogram of one hundred thousand chicks with known sexes and continuously receives the spectrogram uploaded by the local classifier. The network server trains a convolutional neural network by adopting the chick spectrogram with the known gender to obtain a trained network server side classifier (training model); the local classifier does not participate in the training process, and only the conclusion of neural network training in the network server is used, that is, the trained network server side classifier (local classifier) judges spectrogram image signals, and the original local classifier (model) in the processor 2 is downloaded in the network server during the production of the detector.

In the embodiment of the invention, a specific Alexnet network is adopted in the training process of the network server side classifier in the network server, and the training network has 11 layers, wherein the training network comprises 5 convolutional layers, 3 pooling layers and 3 full-connection layers. As can be seen from fig. 5, the parameters of each layer, and the connection sequence of each layer, and the specific operation corresponding to each layer are shown in table 1, which will be explained in detail below.

TABLE 1 operations for each layer

Number of layers

Input layer

Layer

1

Layer 2

Layer 3

Layer 4

Layer 5

Operation of

INPUT

Conv1+RELU

POOL1

Conv2+RELU

POOL2

Conv3+RELU

Number of layers

Layer 6

Layer 7

Layer 8

Layer 9

Layer 10

Layer 11

Operation of

Conv4+RELU

Conv5+RELU

POOL3

FC1+RELU

FC2+RELU

FC3+RELU

As shown in fig. 5, in the stage of model training, spectrogram (picture) of one hundred thousand chicks with known gender is input.

The total number of the convolution layers is 5, taking the first convolution layer as an example: the first convolutional layer has 96 convolution kernels of 11 x 11 with a step size of 4. The convolution operation multiplies the matrix data in fig. 6 one by one using a convolution kernel (in the present embodiment, the convolution kernel is 11 × 11), and then adds them. In the process of extracting the features, the step size of convolution is set (the step size is 4 in this text), the convolution kernel is made to slide from left to right and from top to bottom on the corresponding image, and the convolution operation is performed.

The total number of 3 pooling layers is included, and the main purpose of pooling is to reduce dimension. In the process of extracting abstract features by the convolutional layer, a plurality of convolutional kernels are added, so that the dimension of the extracted features is high, the calculated amount is too large, and overfitting is easy to occur. Taking the first pooling layer in fig. 5 as an example, the filter size is 3 x 3 and the step size is 2 (see fig. 7).

The neural network is somewhat like a black box, the middle identification process is unknown, after convolution and pooling operations are carried out for a plurality of times, abstract feature extraction of data is completed, and a plurality of full connection layers are added at the tail end of the network, so that the classification performance of the network is improved, and classification tasks are completed.

The invention has the beneficial effects that: the detector provided by the invention simplifies the gender detection of the chicks and saves the manual labor; the detection method combines image processing and sound processing technologies, visualizes time domain and frequency domain information in the sound of the chicks, and further analyzes voiceprint characteristics presented in a spectrogram of the chicks to realize classification and identification of the sounding of the chicks, know the gender of the chicks, improve the welfare breeding of the laying hens and improve the production efficiency.

The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents, improvements and the like that fall within the spirit and principle of the present invention are intended to be included therein.

Claims

1. A chick sex detector, comprising: the system comprises an audio collector (1), a processor (2), a display screen (3), a wireless connecting device (4), a power supply (5), a loudspeaker (10), a body (11), function keys (12) and a network server;

the body (11) is in a hollow cylindrical shape, a cavity is formed inside the body, the display screen (3) and the function keys (12) are embedded on the outer surface of the body (11), and the audio collector (1), the processor (2), the wireless connection device (4), the power supply (5) and the loudspeaker (10) are fixedly arranged in the cavity of the body (11);

the power supply (5) is respectively and electrically connected with the audio collector (1), the processor (2), the display screen (3), the wireless connecting device (4) and the loudspeaker (10) to supply power to the whole detector; the audio collector (1), the display screen (3), the wireless connecting device (4), the loudspeaker (10) and the function key (12) are respectively and electrically connected with the processor (2); the processor (2) communicates with the network server through the wireless connection device (4);

the processor (2) receives the trigger signal of the function key (12), and controls the audio collector (1), the display screen (3), the wireless connection device (4) and the loudspeaker (10) respectively according to the trigger signal.

2. The chick sex detector of claim 1, wherein: an audio mounting hole corresponding to the audio collector (1) is formed in one end of the body (11), and the audio collector (1) is mounted at one end inside the body (11) through the audio mounting hole; the other end of the body (11) is provided with a loudspeaker mounting hole corresponding to the loudspeaker (10), and the loudspeaker (10) is fixedly mounted at the other end in the body (11) through the loudspeaker mounting hole (10).

3. The chick sex detector of claim 2, wherein: the audio collector (1) is a microphone, and an audio collecting port faces the outer side of the body (11) through the audio mounting hole so as to collect audio conveniently; the loudspeaker port of the loudspeaker (10) faces the outer side of the body (11) through the loudspeaker mounting hole so as to play audio conveniently.

4. The chick sex detector according to claim 1, characterized in that the processor (2) is a raspberry processor with a memory card, the display screen (3) is an L12864 liquid crystal display screen, and the wireless connection device (4) is an ESP8266 serial port WIFI module.

5. The chick sex detector of claim 1, wherein: the function key (12) comprises: the saving key (6), the clearing key (7), the recording key (8) and the starting key (9) are common mechanical buttons or touch buttons, and are respectively electrically connected with different I/O ports of the processor (2) through signal lines, corresponding keys are pressed down, the processor (2) acquires corresponding trigger signals, and therefore other devices are controlled to respond.

6. The chick sex detector of claim 1, wherein: the power supply (5) is a rechargeable battery with a charging interface; the charging interface is embedded in the outer surface of the body (11) to be connected with an external charging wire to charge the rechargeable battery.

7. A method for detecting the gender of a chick is applied to a chick gender detector; the method is characterized in that: the method for detecting the gender of the chick specifically comprises the following steps:

s101: the audio collector (1) collects the chick singing, obtains a chick singing audio signal and sends the chick singing audio signal to the processor (2);

s102: the processor (2) processes the acquired chick crying sound audio signal to obtain a corresponding two-dimensional spectrogram image signal;

s103: the processor (2) judges the gender of the chick by adopting an Alexnet network algorithm according to the spectrogram image signal to obtain a final judgment result of the gender of the chick;

s104: and the display screen (3) displays the final judgment result of the gender of the chick for viewing.

8. The method for detecting the gender of the chicken as claimed in claim 7, wherein: in step S102, the processor (2) converts the chick squeaking audio signal into a two-dimensional spectrogram image signal by using a spectogram function in matlab.

9. The method for detecting the gender of the chicken as claimed in claim 7, wherein: in the step S103, the processor (2) judges the gender of the chick by adopting an Alexnet network algorithm according to the spectrogram image signal to obtain a chick gender judgment result; the method specifically comprises the following steps:

s201: the processor (2) adopts a local classifier to compare and judge the spectrogram image signals to obtain a preliminary judgment result of the gender of the chick, and the preliminary judgment result is displayed through a display screen (3);

the local classifier is a trained network server classifier obtained after the network server trains the convolutional neural network; the initial local classifier is directly downloaded in a network server by the processor (2) through the wireless connection device (4);

s202: the processor (2) sends the preliminary judgment result and the spectrogram image signal to the network server together through the wireless connection device (4);

s204: the network server sends the final judgment result to the processor (2); the processor (2) judges whether the preliminary judgment result is consistent with the final judgment result; if yes, go to step S205; otherwise, go to step S206;

s205: the processor (2) stores the final judgment result locally for viewing; and to step S207;

s206: the processor (2) stores the final judgment result in the local and updates a local classifier according to the trained network server side classifier in the network server;

s207: and (6) ending.

10. The chick sex detection method of claim 9, wherein in step S203, the convolutional neural network comprises a convolutional layer, a pooling layer and a fully connected layer which are connected in sequence, and an activation function of the convolutional neural network adopts an RE L U function, wherein the convolutional layer comprises 5 convolutional layers, the pooling layer comprises 3 pooling layers, and the fully connected layer comprises 3 fully connected layers.