"Refer to Arduino, I believe that most of the engineers will not be unfamiliar. The current Arduino should not be simple to understand as a single hardware development board, which should also include extensive ecosystem support, open source community, complete project application, etc. A complete open source project, simple and easy to use, so that every electronic enthusiast who likes DIY is fun. Of course, as the hardware is constantly updating iteration, the original Arduino hardware platform is in performance Still restrict, this may be a regrettion of Arduino enthusiasts.
Arduino 101 development board
However, in last year, Arduino Community Joint Intel released the Arduino 101 / Genuino 101 open source hardware board based on Intel Curie module, although two different names, in fact, these two versions are consistent in hardware, different The name and different colors are only for sale in the sale. For example, Arduino 101 is only available in USA, while Genuino 101 is a genuine goods sold in areas other than USA. Compared to the Arduino development board, Arduino 101 / Genuino 101 has improved performance and function, but the most exciting of the author is the INTEL CURIE module on the development board with the neural network memory unit, that is To put it, we can take the development board for deep learning. This is more than a cow, after all, there are not many neuro memory chips that are commercially available, and Intel Curie is just one of them. Fortunately, the third time the author took the Arduino 101, I learned some, I can't wait to share it with you.
This article originally address: https://www.eeboard.com/evaluation/Arduino-101/
Because it is an open source product, the information is found online, so there is only one thanks to the card in addition to the board.
Seriously, the Arduino development board played before, is a treasure spending dozens of money, and the function and genuine although there is no difference, it can clearly see the difference between work and genuine, especially the PCB sheet. The Arduino 101 gave me the biggest feelings in the moment is sophisticated. In addition to some black heart business, in general, the quality is indeed proportional to the price, so if there is an important project, don't be greedy. Small cheap.
A bit embarrassing is that the genuine Arduino 101 is a USB B-type interface, which is not our common MicroSB or Miniusb data cable, so this data / power cord must be purchased separately. In addition, you can see that there are two reset buttons, "Master Reset", "Master Reset", "Master Reset", and "Master Reset" buttons on the Arduino 101 development board USB interface, and the Master Reset is used to reset the entire Intel Curie module, and the reset is used to reset The Arduino program is running. Friends who have been exposed to the Intel Galileo development board are very familiar with this, and it is also provided to provide some APIs for users to decorate in the ARDUINO development.
Relying on the super integration of the onboard Intel Curie module, the board itself is very simple, and the onboard device is also at a glance. Where TI level conversion IC is mainly used for the level conversion of the Intel Curie module and Arduino standard IO.
Arduino development board Main onboard resources:
Intel Curie Module
The core chip on the Arduino 101 board is Intel Curie module. The entire module has only 11mm * 8.05mm size, high 1.95mm, very small, but the integration is not low, an Intel X86 Quark SE microcontroller and a 32bit The core of the Arc architecture (two core sharing flash and ram), NODIC NRF51822 low-power Bluetooth chip, Bosch 6-axis acceleration / gyroscope sensor, multiple LDOs, and DC / DC converters, internal battery charging units, etc. Very powerful, very suitable for both performance and space-restricted wearable equipment and Internet of Things products.
Intel Curie Module Hardware System Box is as follows.
In fact, this part of the content is also mentioned in the Curie Nano evaluation of the panel network, but it is not further refined. Since this article tells the neural network memory function of Arduino 101, it must be more in-depth understanding of this Intel Curie model. The group has already mentioned the microcontroller integrated in this module, and the neural network memory function is in this QUARK SE microcontroller.
Neuron Neurons
128 Neurons (neurons) were integrated in Quark SE, and the core of our article was in this NEURONS. What is the working principle of neurons? What is the advantage?
First, you must know that Neurons is not God, it also needs to learn cognition, so it also has a learning process, as shown below.
For example, we usually see the images, text, videos, hear sounds, etc., these are not cognitiveed by Neurons, these source data need to be collected by sensors, then extract corresponding feature vectors, then Neurons go to learning, cognition, and a restriction in the Quark SE microcontroller is only 128 neurons inside, which means that these feature vectors are limited, and the maximum length cannot exceed 128bytes. This is not, someone has questions, if you go to collect a picture of more than 128pixels to deal with, is there any way? At this time, we don't have to be so dead, you can do secondary or even three times to generate a feature vector that is not more than 128bytes.
The following picture is the workflow chart of Neurons from learning to cognition.
Although Neurons is very impressive, you can understand this, you can make a new born child, nothing will, then you start teach him something, slowly he learns what you teach, It can also be used to raise an anti-three. For a while, after a period of study, you can find new carrots to identify it, it can quickly recognize it from its knowledge base, this is carrot.
But the machine is the machine, or is limited to a large frame, and human beings have unlimited possibilities, so in synthesis, Neurons can do deep learning, but only on simple deep learning, at least on the hardware level It can't be very complicated, regardless of the primary artificial intelligence.
Through the understanding of Neurons, we can simply summarize the advantages of several Neurons, such as
Neurons learn through the program that has already been edited, do not need further programming, you can learn from the line
It only needs to learn very low power consumption.
You can detect new things or abnormal things and report status
The learning cognitive data within Neurons can be copied or transplanted to other places.
Of course, I will further explain how Neurons achieve deep learning with practical examples.
NEURONS functional application
The development environment of Arduino 101 does not have to be introduced, using the Arduino IDE tool, if there is no support package of Arduino 101, you need to install it in the Development Board Manager, as of now, the latest version should be 2.0.2.
Of course, because this will experience the function of Neurons, it also needs the corresponding library support. The NEURONS library Curieneurons is provided by General Vision, divided into free and professional, and the professional version is still charged, selling price of 19 dollars , Say good open source? It seems that open source is limited to free.
Curieneurons library download address: http://www.general-vision.com/software/curieneurons/
From the introduction, the professional version includes all of the full features of the free version, including the source code of the complete read neuron register, especially several particularly powerful algorithms, such as algorithms of the contextual ability through neural network, is a choice L1 or LSUP specification As the algorithm of distance measurement, using a radial basis function or the code of the neighboring rule classification algorithm (of course, some things for CM1K chips, this neural network chip is developed by Chipx, if you want to deepen, Every one is a big topic or a large research project, which can be priced of 19 dollars, or it is worthwhile. However, for the author, the free version is now available.
With the Arduino IDE load the downloaded Curieneurons library, we can use several free applications directly.
Here, the application of the IMU 6 axis accelerometer / gyroscope is selected, collecting related data, learning and cognition of Arduino 101 current motion status and posture. The source code is as follows
#include "" curieimu.h ""
INT AX, AY, AZ; // Accelerometer Values
INT GX, GY, GZ; // GYRMETER VALUES
INT CALIBRATEOFFSETS = 1; // int to determine WHETER CALIBRATION TAKES Place or Not NOT
#include
Curieneurons HNN;
INT CATL = 0; // category to Learn
INT prevcat = 0; // previously recognized category
INT Dist, Cat, NID, NSR, NCOUNT; // Response from the neurons
//
// Variables use for the calculation of the feature vector
//
#define Samplenbr 10 // Number of Samples To Assemble A Vector
#define signalnbr 6 // ax, AY, AZ, GX, GY, GZ
Int raw_vector [samplenbr * signalnbr]; // Vector accumulating the Raw Sensor Data
Byte Vector [Samplenbr * Signalnbr]; // Vector Holding the Pattern to Learn Or Recognize
INT min = 0xffff, maxa = 0, MING = 0xfff, MAXG = 0, Da = 0, DG = 0;
Void setup ()
{
Serial.begin (9600); // Initialize Serial Communication
While (! serial); // Wait for the Serial Port to Open
// Initialize Device
Serial.Println ("INITIALIZING IMU DEVICE ...");
Curieimu.begin ();
// use the code below to calibrate Accel / Gyro Offset Values
IF (CalibrateOffsets == 1)
{
Serial.Println ("" "About to Calibrate. Makess" ");
DELAY (5000);
Serial.Print ("" Starting Gyroscope Calibration and Enabling Offset Compensation ... ");
Curieimu.autocalibrategyrooffset ();
Serial.Println ("DONE");
Serial.Print ("Starting Acceleration Calibration and Enabling Offset Compensation ...");
Curieimu.autocalibrateAccelerometerOffset (X_AXIS, 0);
Curieimu.autocalibrateAccelerometerOffset (Y_AXIS, 0);
Curieimu.autocalibrateAccelerometerOffset (Z_AXIS, 1);
Serial.Println ("DONE");
CurieiMu.setAccelerometerRrange (8);
Curieimu.setgyrorRorange (1000);
}
// initialize the neurons and set a conservation max influnce field
IF (hnn.begin () == 0) serial.print ("\ nneural network is initialized!");
else Serial.print ( "" \ nNeural network is NOT properly connected! "");
hNN.forget (500); // set a conservative Max Influence Field prior to learning
Serial.print ( "" \ n \ nEntering loop ... "");
Serial.print ( "" \ nMove the module vertically or horizontally ... "");
Serial.print ( "" \ ntype 1 + Enter if vertical motion "");
Serial.print ( "" \ ntype 2 + Enter if horizontal motion "");
Serial.print ( "" \ ntype 0 + Enter for any other motion "");
}
void loop ()
{
// Learn if push button depressed and report if a new neuron is committed
//
// wait for a keyboard input of 1 digit between 0-2
// WARNING: make sure the serial printer settings is "" new line ""
//
if (Serial.available () == 2)
{
catL = Serial.read ();
Serial.read (); // to empty serial buffer of the newline char
catL = catL - 48;
if (catL <3) // expected category input (1-vertical, 2-horizontal, 0-still)
{
Serial.print ( "" \ nLearning motion category ""); Serial.print (catL);
//
// learn 5 consecutive sample vectors
// (make sure to keep moving the 101 board accordingly)
//
for (int i = 0; i <5; i ++)
{
extractFeatureVector (); // the vector array is a global
// Optional display of the vector in the serial monitor
//Serial.print(""\nVector = "");
// for (int i = 0; i maxa) maxa = ax; else if (ax maxa) maxa = ay; else if (ay maxa) maxa = az; else if (az maxg) maxg = gx; else if (gx maxg) maxg = gy; else if (gy maxg) maxg = gz; else if (gz 
