Github Repo: https://github.com/codeamt/WindFarmSpotter

Github回购： https : //github.com/codeamt/WindFarmSpotter

This is a series:

这是一个系列：

• Part 1: A Brief Introduction on Leveraging Edge Devices and Embedded AI to Track the U.S.Wind Energy Footprint (You are Here)

  第1部分：有关利用边缘设备和嵌入式AI跟踪USWind能源足迹的简要介绍(您在这里)

• Part 2: An Approach to Satelite Arial Image Data Generation and Automation with Google Earth Engine, Basemap, and Colab

  第2部分： 使用Google Earth Engine，底图和Colab进行卫星Arial图像数据生成和自动化的方法

• Part 3: Experimenting with Memory, Efficiency, and Scaling Input Resolution using a Fast.ai v3 Training Pipeline

  第3部分： 使用Fast.ai v3培训管道试验内存，效率和扩展输入分辨率

• Part 4: Running Inference Tests: Swift-Python Interoperability, and Hardware Acceleration

  第4部分：运行推理测试：Swift-Python互操作性和硬件加速

• Part 5: Spinning Up Inference APIs — Flask (Just Python) v. Kitura (Python & Swift)

  第5部分 ：旋转推理API — Flask(仅Python)诉Kitura(Python和Swift)

• Part 6: Containerizing Deployments for Web, ARMv8/Jetson NVIDIA Series, and SWAP Hardware Platforms

  第6部分： Web，ARMv8 / Jetson NVIDIA系列和SWAP硬件平台的容器化部署

Recently, I completed a data science and software engineering project as part of a hiring pipeline.

最近，我在招聘流程中完成了一个数据科学和软件工程项目。

The company (and I’ll keep the entity anonymous for now) takes a novel approach to the technical interview — lending applicants an NVIDIA Jetson TX2 GPU with free range to execute on a deep learning area of interest.

该公司(我现在将实体保持匿名)将采用一种新颖的方式进行技术面试-向申请人提供具有自由范围的NVIDIA Jetson TX2 GPU，以便在感兴趣的深度学习领域内执行。

关注的领域：风电场—确定潜在的扩展区域，这意味着通过公吨减少碳排放(CO2) (Area of Interest: Wind Farms — Identifying Potential Areas of Expansion Means Reducing Carbon Emission (CO2) by the Metric Ton)

Given the election season and lots of mention of shifting to renewable energy sources being key to lowering our Carbon Footprint, I took this opportunity to learn more about various forms of energy and realized Wind Energy has lots to offer!

鉴于选举季节和降低可再生能源足迹的关键，很多人都提到转向可再生能源，因此我借此机会了解了更多有关各种形式能源的信息，并意识到风能提供了很多！

During my research, I found this fact sheet published by the University of Michigan that laid out the value propositions of Wind Energy. The publication highlighted that:

在研究过程中，我发现了密歇根大学发布的这份情况说明书 ，列出了风能的价值主张。 该出版物强调：

• Increasing Wind Capacity by 1 GigaWatt (GW) avoids the need for Carbon (CO2) Emission by a couple of million metric tons and reduces the need for Water (for Power plants) by roughly a million gallons.
  将风力发电能力提高1吉瓦(GW)，可避免将碳(CO2)排放减少几百万公吨，并减少大约一百万加仑的水(用于发电厂)。

• Previous research from 2015 found that if Wind Turbines — the central technology of Wind Farms — generated 35% of our electricity, this would eliminate 510 billion kg of CO2 emissions annually.

  2015年的先前研究发现，如果风力涡轮机 ( 风力发电场的核心技术)产生了我们35％的电力，那么每年将减少5100亿公斤的二氧化碳排放。

• Wind Farms do not disturb the peace. Given a 350meter radius, Wind Farms emit roughly the same amount of noise (35–45 decibels) as a quiet bedroom (35 decibels) and less noise than a car driving 40mph (55 decibels).

  风电场不会干扰和平。 在半径为350米的情况下，风电场发出的噪音与安静的卧室(35分贝)大致相同(35-45分贝)，并且比以40英里/小时的速度行驶(55分贝)的汽车要少 。

• Wind Energy is very cost-effective. In terms of residential energy prices, in 2016, typical energy quotes were based on the rate of 12.9¢/kWh, where wind energy would only be 2¢/kWh. (That’s right, wind energy would make your electricity bill 6x cheaper!)
  风能非常划算。 在居民能源价格方面，2016年，典型能源报价基于12.9美分/千瓦时的价格，而风能仅为2美分/千瓦时。 (是的，风能会使您的电费便宜6倍！)
• For Wind Farmers, working on large capacity projects (defined in the fact sheet as >= 83 acres), the ROI ratio is $4 to $1.
  对于从事大型项目(在情况说明书中定义为> = 83英亩)的风力发电场，ROI比率为4：1。

Learning about this market has been a whirlwind, to say the least.

至少可以说，了解这个市场是一个旋风。

All this new knowledge made me wonder if data science/deep learning and specifically, computer vision, could help in “spotting potential” regions of interest for new Wind Farm projects and this initial inquiry led to the core idea of my project Wind Farm Spotter: an inference engine for classifying the capacity of existing land-based Wind Farms and potential capacity of unoccupied locations from satellite images.

所有这些新知识使我想知道，数据科学/深度学习，特别是计算机视觉是否可以帮助“发现”新风电场项目的潜在感兴趣区域，而最初的询问导致了我的项目“风电场观测者”的核心思想：推理引擎，用于根据卫星图像对现有陆上风电场的容量和未占用位置的潜在容量进行分类。

项目范围：开发用于风电场观测器的机器学习管道的端到端演练 (Project Scope: An End-to-End Walkthrough of Developing a Machine Learning Pipeline for Wind Farm Spotter)

In subsequent posts, I’ll share my thoughts and findings on developing an end-to-end Machine Learning Pipeline and creating inference engine deployments for web and fog/edge SWAP Hardware Architecture.

在随后的文章中，我将分享我对开发端到端机器学习管道以及为Web和fog / edge SWAP硬件架构创建推理引擎部署的想法和发现。

Tools and Environment:

工具和环境：

Software used to develop this project include:

用于开发此项目的软件包括：

• Google Earth Engine
  Google Earth Engine
• Basemap
  底图
• ArcGIS API Service
  ArcGIS API服务
• PyTorch 1.1 / Torchvision
  PyTorch 1.1 / Torchvision
• pytorchcv
  pytorchcv
• Fast.ai v3
  Fast.ai v3
• Python 3.6, Flask
  Python 3.6，烧瓶
• Swift 5.0.1, Kitura
  雨燕5.0.1，基图拉
• Jetpack 4.3
  喷气背包4.3
• XQuartz (X11)
  XQuartz(X11)
• Virtualenv
  虚拟环境
• Docker Community Edition, Edge
  Docker社区版，Edge

Environment:

环境：

• Google Drive
  Google云端硬碟
• Google Colab
  Google Colab
• MacBook Pro
  MacBook Pro
• Jetson TX2
  杰特逊TX2
• Ubuntu 18.04.3
  Ubuntu 18.04.3

Stay tuned for future posts! The code repository for this series can be found here.

请继续关注以后的帖子！ 该系列的代码存储库可以在这里找到。

Keep Reading:

继续阅读：

Next Post: Part 2: An Approach to Satelite Arial Image Data Generation and Automation with Google Earth Engine, Basemap, and Colab

下一篇文章：第2部分： 使用Google Earth Engine，底图和Colab进行卫星Arial图像数据生成和自动化的方法