# 慈濟大學醫學資訊學系 穿戴式人工智慧工作坊 (2024/05/09) ## 利用TinyML技術快速搭建微型智慧應用(快速指令表) 講師:歐尼克斯實境互動工作室 許哲豪(Jack Hsu)博士    2024/05/09 整理製作 [toc] ## 1. 微型機器學習 ### 1.1 邊緣智慧與生成智慧 ### 1.2 邊緣智慧分級 ### 1.3 智慧物聯網架構 ### 1.4 TinyML 簡介 - [tinyML基金會](https://www.tinyml.org/) - [tinyML Youtube](https://www.youtube.com/@tinyML/videos) ### 1.5 應用、限制、產品、案例 - [TinyML應用大全(30組案例)](https://hackmd.io/@OmniXRI-Jack/tinyML_30_projects) ### 1.6 AIoT 與 TinyML 比較 ## 2. 智慧穿戴式裝置開發流程 ### 2.1智慧穿戴式裝置系統架構 ### 2.2 常見感測技術及市售模組 - [Seeed Studio Grove](https://www.seeedstudio.com/category/Grove-c-1003.html) ### 2.3 常見穿戴式感測器及產品 ### 2.4 感測器類比/數位信號互換 - [【vMaker EDGE AI專欄 #02】 要玩AI前,先來認識數字系統](https://omnixri.blogspot.com/2023/02/vmaker-edge-ai-02-ai.html) ### 2.5 TinyML 應用開發流程 ## 3. TinyML實驗開發板 ### 3.1 實驗器材及開發平台 - [Seeed Wiki - Xiao nRF52840 Sense(英文)](https://wiki.seeedstudio.com/XIAO_BLE/) - [Arduino Software Downloads](https://www.arduino.cc/en/software) - [Edge Impulse Document](https://docs.edgeimpulse.com) ### 3.2 開發板基本介紹 - [Seeed Wiki - Xiao nRF52840 Sense(簡中)](https://wiki.seeedstudio.com/cn/XIAO_BLE/) ### 3.3 Arduino環境建置 #### 3.3.1 連接開發板 以 USB Type C纜線連接電腦和開發板,檢查「裝置管理員」下「連接埠(COM和LPT)」是否有多一個「USB序列裝置(COMxx)」。xx即為埠號,會隨電腦即插入USB位置每次都會隨機配置。若沒有產生序列裝置,則可快速按開發板「Reset」鍵二次,令其進入模式。 #### 3.3.2 安裝 Arduino 開發環境 1. 安裝 Arduino 2.x (Windows Win 10 and newer, 64 bits) https://www.arduino.cc/en/software 2. 開啟Arduino 3. 點選選單 File > Preference ... ,並於 "Additional Boards Manager URLs" 項目填入 **https://files.seeedstudio.com/arduino/package_seeeduino_boards_index.json** 4. 點選選單 Tools > Board > Boards Manager... , 輸入 **nRF52840** 搜尋 Seeed nRF52840 開發板相關函式庫。 安裝下列二個函式庫。版本選最新即可,按下「INSTALL」鍵後開始安裝 4.1. **Seeed nRF52 Boards (BLE, 低功耗用)** 4.2. **Seeed nRF52 mbed-enabled Boards (PDM, IMU, ML使用)** 5. 指定好欲工作的開發板名稱及埠號,點選選單 Tools > Board > Seeed nRF52 Boards > Seeed XAIO nRF52840 Sense 並選擇 Port。可點選 Tools > Get Board Info 來檢查連線是否正確。 ### 3.4 測試程式編譯及上傳(LED閃爍) 1. 點選選單 File > Examples > 01.Basics > Blink 會自動產生下列程式碼 ```c++= // the setup function runs once when you press reset or power the board void setup() { // initialize digital pin LED_BUILTIN as an output. pinMode(LED_BUILTIN, OUTPUT); } // the loop function runs over and over again forever void loop() { digitalWrite(LED_BUILTIN, HIGH); // turn the LED on (HIGH is the voltage level) delay(1000); // wait for a second digitalWrite(LED_BUILTIN, LOW); // turn the LED off by making the voltage LOW delay(1000); // wait for a second } ``` 2. 點選快捷鍵列的**上傳鍵(右箭頭符號)**,此時程式就會開始編譯,完成後會自動上傳到開發板上。如果看到開發板上紅色LED以1秒亮1秒暗,即完成測試。 3. 若無法上傳,快按開發板上RESET鍵二次,再重新選擇開發板及埠號,重新上傳即可。 4. 以上程式 **LED_BUILTIN** (等同於LED_RED)可試著修改成其它顏色LED(紅色LED_RED, 綠色LED_GREEN, 藍色LED_BLUE),或修改延遲時間長度。 ## 4. Edge Impulse Studio ### 4.1 Edge Impulse 簡介 [Edge Impulse](https://edgeimpulse.com/) - [Product](https://edgeimpulse.com/product) - [Public Projects](https://edgeimpulse.com/projects/overview) - [Blog](https://www.edgeimpulse.com/blog/) - [Developer Document](https://docs.edgeimpulse.com/docs) - [Expert network projects](https://edge-impulse.gitbook.io/experts) ### 4.2 建置工作環境 [完整安裝說明](https://docs.edgeimpulse.com/docs/tools/edge-impulse-cli/cli-installation) - 申請並登入 [Edge Impulse 帳號](https://studio.edgeimpulse.com/login) - 安裝 [Python 3.x](https://www.python.org/downloads/) - 安裝 [Node.js (建議安裝v18.20.2 LTS)](https://nodejs.org/en/download) - **安裝 Node.js 時要注意,記得一定要勾選「Tools for Native Modules」** - 以 node -v 及 npm -v 檢查安裝好的版本 - 下載並安裝Arduino CLI (命令列界面) <winodws exe 64bit> [【完整說明】](https://arduino.github.io/arduino-cli/0.35/installation/) - 安裝 Edge Impulse CLI [【完整說明】](https://docs.edgeimpulse.com/docs/tools/edge-impulse-cli/cli-installation) ```npm install -g edge-impulse-cli --force``` - 安裝完成後會得到下列工具 - edge-impulse-daemon - edge-impulse-uploader - edge-impulse-data-forwarder - edge-impulse-run-impulse - edge-impulse-blocks - himax-flash-tool - [Edge Impulse 資料集建置方式](https://edge-impulse.gitbook.io/docs/edge-impulse-studio/data-acquisition) - 從裝置產生(手機、電腦、開發板) - 直接上傳檔案 ### 4.3 啟動公用專案範例 [Edge Impulse Inc. / Tutorial: Recognize sounds from audio](https://studio.edgeimpulse.com/public/14301/latest) 點擊左上角「Clone this project」複製到個人專案,才可開始修改測試。 ## 5. 聲音感測器及喚醒詞偵測 ### 5.1 喚醒詞偵測簡介 ### 5.2 喚醒詞模型原理 ### 5.3 建立 Edge Impulse 專案 Xiao nRF52840 Sense 範例主要參考資料。 - XIAO nRF52840 Sense 上的 PDM 麦克风的使用 https://wiki.seeedstudio.com/cn/XIAO-BLE-Sense-PDM-Usage/ - 基于 Edge Impulse 的语音识别 https://wiki.seeedstudio.com/cn/XIAO-BLE-PDM-EI/ - 基于 TensorFlow Lite 的语音识别 https://wiki.seeedstudio.com/cn/XIAO-BLE-Sense-TFLite-Mic/ **實驗用資料集 on_off_noise_dataset.zip**,包含兩個檔案夾各三個標籤,資料來源 Google speech commands dataset v0.02摘要部份內容。 - train - on(500), off(500), noise(4) - test - on(10), off(10), noise(2) [下載連結](https://drive.google.com/file/d/1awBgM7YAbnhJS3EcWl-vPUpzOgTveA3U/view?usp=sharing) 以下僅提供參考,可以不用下載。 **Google speech commands dataset v0.02** - [完整說明](https://www.tensorflow.org/datasets/catalog/speech_commands) - 訓練集原始壓縮檔 2.37GB, 解壓縮後 3.19GB,36 檔案夾,共 105,841 個檔案。[【手動下載】](http://download.tensorflow.org/data/speech_commands_v0.02.tar.gz) - 測試集原始壓縮檔 110MB, 解壓縮後 149MB,12 檔案夾,共 4,892 個檔案。[【手動下載】](http://download.tensorflow.org/data/speech_commands_test_set_v0.02.tar.gz) #### 5.3.1 建立專案流程 1. 登入帳號,建立新專案,名稱為 Xiao_nRF52840_KWS_Test (名稱可自定),選擇個人專案「Personal」,設定專案屬性為公用「Public」(公用無數量限制,私人限制2個),按下「Create new project」產生新專案。 2. 不連線實體裝置來取樣。 3. 下載實驗用資料集 [on_off_noise_dataset.zip](https://drive.google.com/file/d/1awBgM7YAbnhJS3EcWl-vPUpzOgTveA3U/view?usp=sharing),並解壓縮。 4. 按「+ Add data」,再按「upload data」,進入上傳資料模式。 5. 選擇「Select a folder」,選擇資料集路徑下 \train\on,指定分類為「Training」,指定標籤名稱為「on」,按下左下角「Upload data」完成上傳。 6. 接著依上述步驟5,修改標籤名稱,再分別把「off」和「noise」檔案夾內容上傳。 7. 再來指定分類為「Test」,分別完成上傳資料集路徑\test下的「on」、「off」和「noise」。 8. 由於資料集已完成分割,每個樣本為1秒,所以不需再分割。 9. 接著切換到 Create impulse 項目,修改 Window Size 為 500ms, Increase 100ms, 新增 Audio (MFCC) 及 Classification 區塊,參數使用預值即可,最後按「Save Impulse」儲存設定值。 10. 切換到 Spectral features,切換到 Generate features 子分頁,按下「Generate features」產生對應特徵值。 11. 切換到 Classification,設定訓練次數 30,學習率 0.0005,其它先使用預設值,按下「Start training」即可開始訓練。若推論精度及混淆矩陣表現不理想,可再增加訓練次數。 12. 完成訓練後,可切到 Live Classification 頁面,選擇右半部樣本名稱,再按「Load Sample」即可進行線上測試。 ### 5.4 部署及測試 1. 切換到 Deployment 頁面,選擇 Arduino Library,按下「Build」,即可產生名為「ei-專案名稱-arduino-版本序號.zip」的檔案。 2. 開啟Arduino,點擊選單 Sketch > Include Library > Add .ZIP Library...,將剛才產生的檔案加入。 3. 新增範例檔,點擊選單 File > Examples > ei-專案名稱_inferencing > nano_ble33_sense > nano_ble33_sense_microphone,如下所示。(不包含51-53, 117-134列 新增之程式) 4. 將範例程式直接編譯上傳即可進行推論,此為 No Code 模式。輸出結果以 COM 埠接收文字串。包含推論時間、各標籤置信度。 5. 新增第 51-53 列,定義推論門檻值及LED目前及上一個狀態變數。117-134 列,指定當特定分類置信度高於門檻值時點滅對應的LED。 6. 預設推論時間間隔 2000ms,可自行調整第 86 列程式 delay(n); , n 為時間間隔。 ```c= /* Edge Impulse ingestion SDK * Copyright (c) 2022 EdgeImpulse Inc. * * Licensed under the Apache License, Version 2.0 (the "License"); * you may not use this file except in compliance with the License. * You may obtain a copy of the License at * http://www.apache.org/licenses/LICENSE-2.0 * * Unless required by applicable law or agreed to in writing, software * distributed under the License is distributed on an "AS IS" BASIS, * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. * See the License for the specific language governing permissions and * limitations under the License. * */ // If your target is limited in memory remove this macro to save 10K RAM #define EIDSP_QUANTIZE_FILTERBANK 0 /* ** NOTE: If you run into TFLite arena allocation issue. ** ** This may be due to may dynamic memory fragmentation. ** Try defining "-DEI_CLASSIFIER_ALLOCATION_STATIC" in boards.local.txt (create ** if it doesn't exist) and copy this file to ** `<ARDUINO_CORE_INSTALL_PATH>/arduino/hardware/<mbed_core>/<core_version>/`. ** ** See ** (https://support.arduino.cc/hc/en-us/articles/360012076960-Where-are-the-installed-cores-located-) ** to find where Arduino installs cores on your machine. ** ** If the problem persists then there's not enough memory for this model and application. */ /* Includes ---------------------------------------------------------------- */ #include <PDM.h> #include <Xiao_nRF52840_KWS_Test_inferencing.h> /** Audio buffers, pointers and selectors */ typedef struct { int16_t *buffer; uint8_t buf_ready; uint32_t buf_count; uint32_t n_samples; } inference_t; static inference_t inference; static signed short sampleBuffer[2048]; static bool debug_nn = false; // Set this to true to see e.g. features generated from the raw signal float threshold = 0.50; // 分類置信度門檻值 int LED = 0; // 目前 LED 顯示狀態 int oldLED; // 上一次 LED 顯示狀態 /** * @brief Arduino setup function */ void setup() { // put your setup code here, to run once: Serial.begin(115200); // comment out the below line to cancel the wait for USB connection (needed for native USB) while (!Serial); Serial.println("Edge Impulse Inferencing Demo"); // summary of inferencing settings (from model_metadata.h) ei_printf("Inferencing settings:\n"); ei_printf("\tInterval: %.2f ms.\n", (float)EI_CLASSIFIER_INTERVAL_MS); ei_printf("\tFrame size: %d\n", EI_CLASSIFIER_DSP_INPUT_FRAME_SIZE); ei_printf("\tSample length: %d ms.\n", EI_CLASSIFIER_RAW_SAMPLE_COUNT / 16); ei_printf("\tNo. of classes: %d\n", sizeof(ei_classifier_inferencing_categories) / sizeof(ei_classifier_inferencing_categories[0])); if (microphone_inference_start(EI_CLASSIFIER_RAW_SAMPLE_COUNT) == false) { ei_printf("ERR: Could not allocate audio buffer (size %d), this could be due to the window length of your model\r\n", EI_CLASSIFIER_RAW_SAMPLE_COUNT); return; } } /** * @brief Arduino main function. Runs the inferencing loop. */ void loop() { ei_printf("Starting inferencing in 2 seconds...\n"); delay(1000); // 調整二次取樣推論間隔時間(ms) ei_printf("Recording...\n"); bool m = microphone_inference_record(); if (!m) { ei_printf("ERR: Failed to record audio...\n"); return; } ei_printf("Recording done\n"); signal_t signal; signal.total_length = EI_CLASSIFIER_RAW_SAMPLE_COUNT; signal.get_data = &microphone_audio_signal_get_data; ei_impulse_result_t result = { 0 }; EI_IMPULSE_ERROR r = run_classifier(&signal, &result, debug_nn); if (r != EI_IMPULSE_OK) { ei_printf("ERR: Failed to run classifier (%d)\n", r); return; } // print the predictions ei_printf("Predictions "); ei_printf("(DSP: %d ms., Classification: %d ms., Anomaly: %d ms.)", result.timing.dsp, result.timing.classification, result.timing.anomaly); ei_printf(": \n"); for (size_t ix = 0; ix < EI_CLASSIFIER_LABEL_COUNT; ix++) { ei_printf(" %s: %.5f\n", result.classification[ix].label, result.classification[ix].value); //lets light up some LEDS if (result.classification[ix].value > threshold) { //now let's see what label were in switch (ix) { case 0: LED = 11; break; // RED LED case 1: LED = 12; break; // BLUE LED case 2: LED = 13; break; // GREEN LED default: LED = 0; } //in Sense, LOW will light up the LED if (LED != 0) { digitalWrite (oldLED, HIGH); //if we enter a word right next to previous - we turn off the previous LED digitalWrite (LED, LOW); oldLED = LED; } else //turn off LED digitalWrite (oldLED, HIGH); } } #if EI_CLASSIFIER_HAS_ANOMALY == 1 ei_printf(" anomaly score: %.3f\n", result.anomaly); #endif } /** * @brief PDM buffer full callback * Get data and call audio thread callback */ static void pdm_data_ready_inference_callback(void) { int bytesAvailable = PDM.available(); // read into the sample buffer int bytesRead = PDM.read((char *)&sampleBuffer[0], bytesAvailable); if (inference.buf_ready == 0) { for(int i = 0; i < bytesRead>>1; i++) { inference.buffer[inference.buf_count++] = sampleBuffer[i]; if(inference.buf_count >= inference.n_samples) { inference.buf_count = 0; inference.buf_ready = 1; break; } } } } /** * @brief Init inferencing struct and setup/start PDM * * @param[in] n_samples The n samples * * @return { description_of_the_return_value } */ static bool microphone_inference_start(uint32_t n_samples) { inference.buffer = (int16_t *)malloc(n_samples * sizeof(int16_t)); if(inference.buffer == NULL) { return false; } inference.buf_count = 0; inference.n_samples = n_samples; inference.buf_ready = 0; // configure the data receive callback PDM.onReceive(&pdm_data_ready_inference_callback); PDM.setBufferSize(4096); // initialize PDM with: // - one channel (mono mode) // - a 16 kHz sample rate if (!PDM.begin(1, EI_CLASSIFIER_FREQUENCY)) { ei_printf("Failed to start PDM!"); microphone_inference_end(); return false; } // set the gain, defaults to 20 PDM.setGain(127); return true; } /** * @brief Wait on new data * * @return True when finished */ static bool microphone_inference_record(void) { inference.buf_ready = 0; inference.buf_count = 0; while(inference.buf_ready == 0) { delay(10); } return true; } /** * Get raw audio signal data */ static int microphone_audio_signal_get_data(size_t offset, size_t length, float *out_ptr) { numpy::int16_to_float(&inference.buffer[offset], out_ptr, length); return 0; } /** * @brief Stop PDM and release buffers */ static void microphone_inference_end(void) { PDM.end(); free(inference.buffer); } #if !defined(EI_CLASSIFIER_SENSOR) || EI_CLASSIFIER_SENSOR != EI_CLASSIFIER_SENSOR_MICROPHONE #error "Invalid model for current sensor." #endif ``` ## 6. 運動感測器及手勢辨識 ### 6.1 運動感測簡介 ### 6.2 IMU 取值及輸出 1. 至 [Seeed Arduino LSM6DS3 Github](https://github.com/Seeed-Studio/Seeed_Arduino_LSM6DS3) 下載函式庫(*.zip)。 2. 開啟 Arduino IDE,點擊選單 Sketch > Include Library > Add .ZIP Library... 新增下載的函式庫。 3. 點擊選單 File > Examples > Seeed Arduino LSM6DS3 > IMU_Capture 開啟範例。 4. 依下列指示修改範例程式,方便後續連續輸出,可搭配 edge-impulse-data-forward 將感測器數值送到雲端。 - 加入 // 註解掉第 7-9 列 - 修改第 13 列,將COM埠通訊速度提升到 115,200 bps - 加入 // 註解掉第 25-49 列及第 65-69 列,讓程式以最快方式輸出 IMU 加速度計及陀螺儀數值。 5. 編譯並上傳程式到開發板,可點擊選單 Tools > Serial Monitor 或 Serial Plotter 監看 IMU 輸出數值文字串或數值圖形內容。 ```c= #include <LSM6DS3.h> #include <Wire.h> //Create a instance of class LSM6DS3 LSM6DS3 myIMU(I2C_MODE, 0x6A); //I2C device address 0x6A float aX, aY, aZ, gX, gY, gZ; // const float accelerationThreshold = 2.5; // threshold of significant in G's // const int numSamples = 119; // int samplesRead = numSamples; void setup() { // put your setup code here, to run once: Serial.begin(115200); while (!Serial); //Call .begin() to configure the IMUs if (myIMU.begin() != 0) { Serial.println("Device error"); } else { Serial.println("aX,aY,aZ,gX,gY,gZ"); } } void loop() { // wait for significant motion // while (samplesRead == numSamples) { // // read the acceleration data // aX = myIMU.readFloatAccelX(); // aY = myIMU.readFloatAccelY(); // aZ = myIMU.readFloatAccelZ(); // // sum up the absolutes // float aSum = fabs(aX) + fabs(aY) + fabs(aZ); // // check if it's above the threshold // if (aSum >= accelerationThreshold) { // // reset the sample read count // samplesRead = 0; // break; // } // } // // check if the all the required samples have been read since // // the last time the significant motion was detected // while (samplesRead < numSamples) { // // check if both new acceleration and gyroscope data is // // available // // read the acceleration and gyroscope data // samplesRead++; // print the data in CSV format Serial.print(myIMU.readFloatAccelX(), 3); Serial.print(','); Serial.print(myIMU.readFloatAccelY(), 3); Serial.print(','); Serial.print(myIMU.readFloatAccelZ(), 3); Serial.print(','); Serial.print(myIMU.readFloatGyroX(), 3); Serial.print(','); Serial.print(myIMU.readFloatGyroY(), 3); Serial.print(','); Serial.print(myIMU.readFloatGyroZ(), 3); Serial.println(); // if (samplesRead == numSamples) { // // add an empty line if it's the last sample // Serial.println(); // } // } } ``` ### 6.3 建立 Edge Impulse 專案 Xiao nRF52840 Sense 範例主要參考資料。 - XIAO nRF52840 Sense 上的 6-Axis IMU 的使用 https://wiki.seeedstudio.com/cn/XIAO-BLE-Sense-IMU-Usage/ - XIAO nRF52840 Sense 基于 Edge Impulse 的运动识别 https://wiki.seeedstudio.com/cn/XIAOEI/ - 基于 TensorFlow Lite 的运动识别 https://wiki.seeedstudio.com/cn/XIAO-BLE-Sense-TFLite-Getting-Started/ 1. 登入帳號,建立新專案,名稱為 Xiao_nRF52840_IMU_Test (名稱可自定),選擇個人專案「Personal」,設定專案屬性為公用「Public」(公用無數量限制,私人限制2個),按下「Create new project」產生新專案。 2. 使用開發板連線方式來取樣及測試。開啟命令列模式執行「``` edge-impulse-data-forwarder --clean```」,接著輸入帳號、密碼,按上下鍵選擇對應的專案名稱(Xiao_nRF52840_IMU_Test),輸入開發板名稱(可自由命名)及資料數量及名稱(用逗號分隔),此時應可順利連接開發板到雲端,可打開 Edge Impulse 專案 Devices 檢查是否綠燈。 3. 進入 Edge Impulse Data acquistion 右半部 Data Collect data,選擇裝置,指定標籤名稱,設定取樣時長(10000ms),按下「Start sampling」開始取樣。開始依定義手勢重覆執行多次(中間要有間隔),直到時間結束。 4. 點擊新增資料末端三小點符號,選擇「Split Sample」,進入分割資料模式,將每個樣本為1秒。 5. 進入後會自動設定分割位置及數量,如果不滿意,可自行調整,最後按下「Split」即可自動將內容分解成多個樣本。預設全部加入訓練集中。 6. 接著切換到 Create impulse 項目,修改 Window Size 為 500ms, Increase 100ms, 新增 Spectral Analysis,為加速訓練時間可僅勾選三軸加速度計(aX, aY, aZ)即可,再新增 Classifier 區塊,參數使用預值即可,最後按「Save Impulse」儲存設定值。 7. 切換到 Spectral features,切換到 Generate features 子分頁,按下「Generate features」產生對應特徵值。 8. 切換到 Classifier,設定訓練次數 30,學習率 0.0005,其它先使用預設值,按下「Start training」即可開始訓練。若推論精度及混淆矩陣表現不理想,可再增加訓練次數。 9. 若效果仍不理想,可將 Neural network architecture 二層 Dense Layer Neural數量都增加到 40(層數、神經元數可自定),再重新訓練即可。 10. 完成訓練後,可切到 Live Classification 頁面,選擇左半部 Classify new data,再按「Start sampling」即可進行取樣並進行線上測試。11. ### 6.4 部署及測試 1. 切換到 Deployment 頁面,選擇 Arduino Library,按下「Build」,即可產生名為「ei-專案名稱-arduino-版本序號.zip」的檔案。 2. 開啟Arduino,點擊選單 Sketch > Include Library > Add .ZIP Library...,將剛才產生的檔案加入。 3. 新增範例檔,點擊選單 File > Examples > ei-專案名稱_inferencing > nano_ble33_sense > nano_ble33_sense_acclerometer,如下所示。(不包含20, 21, 26, 100-102列 新增之程式) 4. 確認開發板設定,點擊選單 Tools > Boards > Seeed nRF52 mbed-enabled Boards > Seeed XIAO BLE Sense - nRF52840 。 5. 由於目前使用的 IMU 為 LSM6DS3 所以需修改程式內容如下。 - 註解掉第 19 列, //#include <Arduino_LSM9DS1.h> - 新增第 20 列, #include <LSM6DS3.h> - 新增第 21 列, #include <Wire.h> - 新增第 26 列, LSM6DS3 myIMU(I2C_MODE, 0x6A); - 修改第 56 列, if (!myIMU.begin()) { - 註解掉第 99 列, // IMU.readAcceleration(buffer[ix], buffer[ix + 1], buffer[ix + 2]); - 新增 100-102 列, buffer[ix] = myIMU.readFloatAccelX(); buffer[ix + 1] = myIMU.readFloatAccelY(); buffer[ix + 2] = myIMU.readFloatAccelZ(); 6. 將程式進行編譯後,第 148 列「#error "Invalid model for current sensor"」會產生「"invalid model for current sensor"」錯誤。主要原因為Edge Impulse打包時是以Sensor_Fusion方式處理而不是Accelerometer方式造成。 7. 點擊選單 File > Preferences 找到 Sketchbook Folder,進到該路徑後,進到 \libraries\專案名稱_inferencing\src\model-parameters\ ,開啟model_metadata.h ,找到 「 #define EI_CLASSIFIER_SENSOR EI_CLASSIFIER_SENSOR_FUSION 」,將定義改成「 EI_CLASSIFIER_SENSOR_ACCELEROMETER 」存檔即可。 8. 將範例程式直接編譯上傳即可進行推論,此為 No Code 模式。輸出結果以 COM 埠接收文字串。包含推論時間、各標籤置信度。 9. 預設推論時間間隔 2000ms,可自行調整第 88 列程式 delay(n); , n 為時間間隔。 10. 如需驅動LED或其它硬體,可於第 140列後加入。 ```c= /* Edge Impulse ingestion SDK * Copyright (c) 2022 EdgeImpulse Inc. * * Licensed under the Apache License, Version 2.0 (the "License"); * you may not use this file except in compliance with the License. * You may obtain a copy of the License at * http://www.apache.org/licenses/LICENSE-2.0 * * Unless required by applicable law or agreed to in writing, software * distributed under the License is distributed on an "AS IS" BASIS, * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. * See the License for the specific language governing permissions and * limitations under the License. * */ /* Includes ---------------------------------------------------------------- */ #include <Xiao_nRF52840_IMU_Test_inferencing.h> //#include <Arduino_LSM9DS1.h> //Click here to get the library: https://www.arduino.cc/reference/en/libraries/arduino_lsm9ds1/ #include <LSM6DS3.h> #include <Wire.h> /* Constant defines -------------------------------------------------------- */ #define CONVERT_G_TO_MS2 9.80665f #define MAX_ACCEPTED_RANGE 2.0f // starting 03/2022, models are generated setting range to +-2, but this example use Arudino library which set range to +-4g. If you are using an older model, ignore this value and use 4.0f instead LSM6DS3 myIMU(I2C_MODE, 0x6A); //I2C device address 0x6A /* ** NOTE: If you run into TFLite arena allocation issue. ** ** This may be due to may dynamic memory fragmentation. ** Try defining "-DEI_CLASSIFIER_ALLOCATION_STATIC" in boards.local.txt (create ** if it doesn't exist) and copy this file to ** `<ARDUINO_CORE_INSTALL_PATH>/arduino/hardware/<mbed_core>/<core_version>/`. ** ** See ** (https://support.arduino.cc/hc/en-us/articles/360012076960-Where-are-the-installed-cores-located-) ** to find where Arduino installs cores on your machine. ** ** If the problem persists then there's not enough memory for this model and application. */ /* Private variables ------------------------------------------------------- */ static bool debug_nn = false; // Set this to true to see e.g. features generated from the raw signal /** * @brief Arduino setup function */ void setup() { // put your setup code here, to run once: Serial.begin(115200); // comment out the below line to cancel the wait for USB connection (needed for native USB) while (!Serial); Serial.println("Edge Impulse Inferencing Demo"); if (!myIMU.begin()) { ei_printf("Failed to initialize IMU!\r\n"); } else { ei_printf("IMU initialized\r\n"); } if (EI_CLASSIFIER_RAW_SAMPLES_PER_FRAME != 3) { ei_printf("ERR: EI_CLASSIFIER_RAW_SAMPLES_PER_FRAME should be equal to 3 (the 3 sensor axes)\n"); return; } } /** * @brief Return the sign of the number * * @param number * @return int 1 if positive (or 0) -1 if negative */ float ei_get_sign(float number) { return (number >= 0.0) ? 1.0 : -1.0; } /** * @brief Get data and run inferencing * * @param[in] debug Get debug info if true */ void loop() { ei_printf("\nStarting inferencing in 2 seconds...\n"); delay(1000); ei_printf("Sampling...\n"); // Allocate a buffer here for the values we'll read from the IMU float buffer[EI_CLASSIFIER_DSP_INPUT_FRAME_SIZE] = { 0 }; for (size_t ix = 0; ix < EI_CLASSIFIER_DSP_INPUT_FRAME_SIZE; ix += 3) { // Determine the next tick (and then sleep later) uint64_t next_tick = micros() + (EI_CLASSIFIER_INTERVAL_MS * 1000); // IMU.readAcceleration(buffer[ix], buffer[ix + 1], buffer[ix + 2]); buffer[ix] = myIMU.readFloatAccelX(); buffer[ix + 1] = myIMU.readFloatAccelY(); buffer[ix + 2] = myIMU.readFloatAccelZ(); for (int i = 0; i < 3; i++) { if (fabs(buffer[ix + i]) > MAX_ACCEPTED_RANGE) { buffer[ix + i] = ei_get_sign(buffer[ix + i]) * MAX_ACCEPTED_RANGE; } } buffer[ix + 0] *= CONVERT_G_TO_MS2; buffer[ix + 1] *= CONVERT_G_TO_MS2; buffer[ix + 2] *= CONVERT_G_TO_MS2; delayMicroseconds(next_tick - micros()); } // Turn the raw buffer in a signal which we can the classify signal_t signal; int err = numpy::signal_from_buffer(buffer, EI_CLASSIFIER_DSP_INPUT_FRAME_SIZE, &signal); if (err != 0) { ei_printf("Failed to create signal from buffer (%d)\n", err); return; } // Run the classifier ei_impulse_result_t result = { 0 }; err = run_classifier(&signal, &result, debug_nn); if (err != EI_IMPULSE_OK) { ei_printf("ERR: Failed to run classifier (%d)\n", err); return; } // print the predictions ei_printf("Predictions "); ei_printf("(DSP: %d ms., Classification: %d ms., Anomaly: %d ms.)", result.timing.dsp, result.timing.classification, result.timing.anomaly); ei_printf(": \n"); for (size_t ix = 0; ix < EI_CLASSIFIER_LABEL_COUNT; ix++) { ei_printf(" %s: %.5f\n", result.classification[ix].label, result.classification[ix].value); } #if EI_CLASSIFIER_HAS_ANOMALY == 1 ei_printf(" anomaly score: %.3f\n", result.anomaly); #endif } #if !defined(EI_CLASSIFIER_SENSOR) || EI_CLASSIFIER_SENSOR != EI_CLASSIFIER_SENSOR_ACCELEROMETER #error "Invalid model for current sensor" #endif ``` ## 學習資源 [1] 歐尼克斯實境互動工作室,TinyML(MCU AI)系列發文 https://hackmd.io/@OmniXRI-Jack/series_articles#TinyMLMCU-AI%E7%B3%BB%E5%88%97 [2] CAVEDU教育團隊技術部落格,TinyML https://blog.cavedu.com/?s=tinyml [3] 【iCShop開箱趣】ep7 Seeed Studio XIAO nRF52840 Sense Review | 把神經網路放到MCU原來這麼簡單? TinyML Tutorial 範例教學 https://youtu.be/Faka_ahto0o [4] 林士允(Felix Lin), 【Maker 玩 AI】 XIAO nRF52840 Sense-入門 TinyML 就靠它! https://vmaker.tw/archives/65534 ## 參考文獻 [1] 許哲豪,NTUST Edge AI 人工智慧與邊緣運算實務 111 學年度 - 7.6.喚醒詞偵測, 7.7.運動手勢辨識 https://github.com/OmniXRI/NTUST_EdgeAI_2023 [2] 許哲豪,【課程簡報】20231222_慈濟醫資_智慧物聯網(AIoT)與智慧醫療應用 https://omnixri.blogspot.com/2023/12/20231222aiot.html [3] 許哲豪,【課程簡報】20231208_南開科大_微型人工智慧應用與實作(基礎篇) https://omnixri.blogspot.com/2023/12/20231208.html [4] 許哲豪,TinyML應用大全(30組案例) https://hackmd.io/@OmniXRI-Jack/tinyML_30_projects [5] 矽递科技(Seeed Studio) Wiki 文档平台 - XIAO nRF52840 (Sense) 开发板 https://wiki.seeedstudio.com/cn/XIAO_BLE/ [6] Edge Impulse, Developers Document https://docs.edgeimpulse.com/ ## 延伸閱讀 [A] 許哲豪,爭什麼,把AI和MCU摻在一起做tinyML就對了! https://omnixri.blogspot.com/2021/10/ithome-2021-13-arm-platforms-aimcutinyml.html [B] 許哲豪,AIoT與tinyML生態系國際發展趨勢與國產IC未來方向 https://omnixri.blogspot.com/2022/11/20221110icaiottinyml-ic.html [C] 許哲豪,MCU攜手NPU讓tinyML邁向新里程碑 https://omnixri.blogspot.com/2022/10/mcunputinyml.html [D] 許哲豪,有了TinyML加持MCU也能開始玩電腦視覺了 https://omnixri.blogspot.com/2022/12/tinymlmcu.html [E] 許哲豪,TinyML 核心函式庫 Arm CMSIS 6 DSP & NN 更新比較 https://omnixri.blogspot.com/2024/02/tinyml-arm-cmsis-6-dsp-nn.html [F] 【vMaker Edge AI專欄 #14】 從CES 2024 看Edge AI及TinyML最新發展趨勢 https://omnixri.blogspot.com/2024/02/vmaker-edge-ai-14-ces-2024-edge-aitinyml.html [G] 許哲豪(史蒂芬周), tinyML這個球不是這麼踢滴系列 https://ithelp.ithome.com.tw/users/20141396/ironman/5860 [H] 許哲豪,TinyML(MCU AI)系列發文 https://hackmd.io/@OmniXRI-Jack/series_articles#TinyMLMCU-AI%E7%B3%BB%E5%88%97