1. A multisensor multi-environment monitoring method based on AIot and TinyML technology is characterized by comprising the following steps:

s1, performing classified collection on data acquired from sensors in different environments;

s2, performing statistical characteristic analysis on data collected by the sensor, and performing influence coefficients of different characteristics on the environment in a scene environment;

s3, training the model by using the characteristics with correlation through a deep learning framework;

s4, converting the obtained model file into a TF Lite format, then performing model curing on the model in the TF Lite format in a binary format, and converting the model into a cpp binary model file and a h model header file;

s5, deploying and reasoning a model according to MCU development boards of different end devices;

s6, reasoning is carried out according to data input by different sensors, and then the next step of adjustment is carried out to maintain the stability of the environment;

and S7, transmitting the data collected by different sensors to a cloud end through a network.

2. The multi-sensor multi-environment monitoring method based on AIot and TinyML technology as claimed in claim 1, wherein: expressing the correlation degree of different pairing characteristics by using the Pearson coefficient, and the formula isWherein x and y are respectively data of different characteristics, N is the total characteristic quantity, and r is a Pearson coefficient.

3. The multi-sensor multi-environment monitoring method based on AIot and TinyML technology as claimed in claim 1, wherein: and (4) training by using an ANN or GRNN model, verifying the model by MSE, and finally obtaining a model file.

4. The multi-sensor multi-environment monitoring method based on AIot and TinyML technology as claimed in claim 1, wherein: the Nano33 BLE Sense of Arduino performs model deployment and reasoning, wherein a TFLite computational library is introduced using the IDE of Arduino, and the resulting cc file is deployed as a reasoning model into the developed application.

Background

Environmental monitoring refers to the activities of environmental monitoring mechanisms to monitor and measure environmental quality conditions. The environmental monitoring is to monitor and measure the index reflecting the environmental quality to determine the environmental pollution condition and the environmental quality. Especially, the monitoring of the water environment is important.

The traditional water environment monitoring and forecasting method is a method combining manpower with monitoring points and sections. The traditional monitoring method has three major defects, firstly, manual sampling is greatly limited by terrain and weather conditions, monitoring dead points, bad weather or sudden events are easy to occur in areas where manual sampling is difficult to reach, and the analysis time period of a manual sampling laboratory is long; secondly, the manual sampling and transportation cost is high, and a large amount of manpower, traffic and expenses are needed; finally, the time period of manual back-and-forth sampling and laboratory analysis is long, so that water quality parameters of the water sample are easy to change, and data errors are caused.

In order to solve the specific problems, people try to develop and build a dynamic monitoring and early warning system of water environment parameters by using the technology of internet of things, overcome the defects of the traditional method and perform water environment protection and dynamic monitoring. However, the existing equipment not only has a complex structure, but also has a single function.

Artificial Intelligence (AI) and internet of things (IoT) are popular areas in computer science. AIoT (artificial intelligence internet of things) is implemented when it fuses AI and IoT together, applying AI to IoT, and these internet of things systems are able to analyze data without human intervention and have decision-making potential.

TinyML is a cross-domain of machine learning and embedded IoT devices, an emerging engineering discipline, with the potential to revolutionize many industries. It is necessary to combine and apply artificial intelligence thing networking and lightweight machine learning to environmental monitoring.

Disclosure of Invention

The invention provides a multi-sensor multi-environment monitoring method based on AIot and TinyML technologies, which realizes artificial intelligence related reasoning of light-weight MCU + sensor equipment, monitors the environment and maintains the stability of the environment.

In order to achieve the purpose, the invention is realized by the following technical scheme:

a multisensor multi-environment monitoring method based on AIot and TinyML technology includes the following steps:

s1, performing classified collection on data acquired from sensors in different environments;

s2, performing statistical characteristic analysis on data collected by the sensor, and performing influence coefficients of different characteristics on the environment in a scene environment;

s3, training the model by using the characteristics with correlation through a deep learning framework;

s4, converting the obtained model file into a TF Lite format, then performing model curing on the model in the TF Lite format in a binary format, and converting the model into a cpp binary model file and a h model header file;

s5, deploying and reasoning a model according to MCU development boards of different end devices;

s6, reasoning is carried out according to data input by different sensors, and then the next step of adjustment is carried out to maintain the stability of the environment;

and S7, transmitting the data collected by different sensors to a cloud end through a network.

On the basis of the multi-sensor multi-environment monitoring method based on the AIot and TinyML technology, the correlation degree of different pairing characteristics is expressed by utilizing the Pearson coefficient, and the formula isWherein x and y are respectively data of different characteristics, N is the total characteristic quantity, and r is a Pearson coefficient.

On the basis of the multi-sensor multi-environment monitoring method based on the AIot and TinyML technologies, an ANN or GRNN model is used for training, the model is verified through MSE, and finally a model file is obtained.

On the basis of the multi-sensor multi-environment monitoring method based on the AIot and TinyML technologies, model deployment and reasoning are carried out by the Nano33 BLE Sense of Arduino, wherein a TFLite calculation library is introduced by utilizing the IDE of Arduino, and then the obtained cc file is deployed into developed application as a reasoning model.

The invention has the advantages that: different types of sensors collect different types of historical data of different environments, feature analysis is carried out on the basis of the collected data, then machine learning or deep learning model training is carried out on related features, AIoT and TinyML technologies are combined, artificial intelligence related reasoning is carried out on the light-weight MCU + sensor equipment, and the environments are monitored and kept stable.

Drawings

The accompanying drawings, which are included to provide a further understanding of the invention and are incorporated in and constitute a part of this specification, illustrate embodiments of the invention and together with the description serve to explain the principles of the invention and not to limit the invention.

Fig. 1 is a block diagram of the structure of the embodiment of the present invention.

Fig. 2 is a schematic visualization diagram according to an embodiment of the present invention.

Detailed Description

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.

A multi-sensor multi-environment detection method based on AIot and TinyML technology. The method mainly comprises modules of data collection, data analysis, deep learning model training, model lightweight, model deployment, reasoning and the like, and specifically comprises the following module contents and steps:

s1: the sensors collect data:

in different environments, classified collection is carried out through data acquired from sensors, and in the embodiment, chemical sensors are used for detecting the components of substances in water;

s2: data characteristic analysis:

carrying out statistical characteristic analysis on data collected by the sensor, and carrying out influence coefficients of different characteristics on the environment in a scene environment, namely carrying out the degree of influence of different substances on water quality by utilizing the water quality component information collected by the sensor;

the correlation degree of different pairing characteristics, namely the correlation relation between different substances and water quality, is expressed by utilizing the Pearson coefficient, and the formula isWherein x represents substances in water, y represents the quality of water, N is total characteristic quantity, r is Pearson coefficient, the value range is 0-1, wherein the larger r represents the stronger correlation between different substances and water quality;

s3: training of the model:

training a model by utilizing a deep learning framework (Tensorflow) for characteristics with strong correlation between substances and water quality, wherein an ANN (Artificial Neural network) or GRNN (general Regression Neural network) model can be utilized for training, the substances are required to be labeled firstly, then the substances with different labels are in one-to-one correspondence with the corresponding water quality, then the deep learning framework is utilized for carrying out Regression training, finally, an error rate is obtained, then, MSE is calculated through the error rate, and the MSE (mean Square error) is utilized for carrying out model verification to optimize the model, and finally, a model file is obtained;

s4: model file conversion:

and carrying out TF Lite format conversion on the obtained model file, then carrying out binary format model solidification on the TF Lite format model, and converting the TF Lite format model into a cpp binary model file and an h model header file. In this process, an xxd-i model may be utilized, tflite > person _ detect _ model _ data.cc instructions for conversion;

s5: and (3) deploying the MCU end model:

deploying and reasoning the model according to MCU development boards of different end devices, such as Nano33 BLE Sense of Arduino, wherein a TFLite calculation library is introduced by using IDE of Arduino, and then the obtained cc file is deployed into the developed application as a reasoning model;

step 6: the MCU processes the sensor data to reason:

reasoning is carried out according to data input by different sensors, in the example, the sensors can continuously monitor the water quality, meanwhile, a deployed model can also reason the input data, and once the water quality is found to exceed a normal value, the next step of adjustment is carried out to maintain the stability of the water quality;

and 7: MCU control equipment maintains ambient temperature

Data collected by different sensors are transmitted to the cloud end through a network (Wifi or 5G) to be visualized, so that workers can monitor the data in real time, and the effect of double insurance is achieved.

The working principle of the invention is as follows: the method comprises the steps of conducting data collection of different sensors through determination of a selected environment, then conducting model training through an artificial intelligence machine learning or deep learning model, then conducting lightweight of the model so as to be conveniently deployed on lightweight MCU + sensor equipment, finally conducting reasoning through input of sensor data, and controlling different equipment through the MCU to maintain stability of the environment. Meanwhile, data of the sensor can be transmitted to the cloud end through the network, and visualization of the data is carried out.

Finally, it should be noted that: although the present invention has been described in detail with reference to the foregoing embodiments, it will be apparent to those skilled in the art that changes may be made in the embodiments and/or equivalents thereof without departing from the spirit and scope of the invention. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.