Using convolutional networks and satellite imagery to identify patterns in urban environments at a large scale

速读 deep learning 在城市应用
Urban planning applications (energy audits, investment, etc.) require an understanding of built infrastructure and its environment,
i.e., both low-level, physical features (amount of vegetation, building area and geometry etc.), as well as higher-level concepts such as land use classes (which encode expert understanding of socioeconomic end uses).

  • 微观层面 和宏观层面 看法不同
  • imageNet transfer learning 过来到自己的数据上。已经有了label

Benchmark 数据集再调试模型的时候很有用。文章提供两个数据集还是很有用。(Existing land use benchmark datasets)

  • UC Merced. is dataset was published in 2010 and contains 2100 256 × 256, 1m/px aerial RGB images over 21 land use classes. It is considered a “solved problem”, as modern neural network based classifiers have achieved > 95% accuracy on it.

  • DeepSat.DeepSat dataset1 was released in 2015. It contains two benchmarks: the Sat-4 data of 500, 000 images over 4 land use classes (barren land, trees, grassland, other), and the Sat-6 data of 405, 000 images over 6 land use classes (barren land, trees, grassland, roads, buildings, water bodies). All the samples are 28×28 in size at a 1m/px spatial resolution and contain 4 channels (red, green, blue, and NIR - near infrared). Currently less than two years old, this dataset is already a “solved problem”, with previous studies (and our own experiments) achieving classification accuracies

Spatio-temporal patterns of soil organic carbon and pH in relation to environmental factors—A case study of the Black Soil Region of Northeastern China - ScienceDirect

速读 结论
The principal factors impacting SOC and pH included: precipitation, gully density, forested land and grain yield. There was significant covariation between natural and human factors in forming these spatial patterns. Anthropogenic disturbance had a larger influence on the distribution of SOC than on the distribution of pH.


Rethinking the Inception Architecture for Computer Vision


5x5 比3x3 快了2.78倍,但是用两个2x2 代替3x3只减少了11%的工作量

一个 feature map 通过一个卷积计算 所谓的共享权重和偏置项

两个3x3 代替一个5x5 可以理解为在一个5x5的区域先用一个3x3 再接着一个fc.

Fine crop mapping by combining high spectral and high spatial resolution remote sensing data in complex heterogeneous areas - ScienceDirect

> 速读


  1. 仅使用高空间分辨率或仅高光谱分辨率的遥感数据很难高精度地绘制作物。
  2. 由于高空间分辨率数据的光谱分辨率太低,在高空间分辨率数据中,不同植被类型的光谱分辨率非常小。使用高空间分辨率数据很难区分不同的植被类型。
  3. 对于高光谱分辨率的遥感数据,由于这些数据的空间分辨率低,很难排除作物中的道路,桥梁和排水沟等线状物体。
  4. 针对这一问题,提出并结合江苏省苏州市将高空域和高光谱分辨率遥感数据结合用于异构地区的基于对象的精细作物制图方法。首先,纯农作物多边形是从0.5米的土地数据中获得的。由于空间分辨率高,非耕地可以很容易地从耕地中分离出来。然后,使用Hyperion数据为每个纯农作物多边形的作物分类。
  5. 结果表明,该方法可以对复杂异构地区的作物进行地图定位,总体精度高于95%,远高于仅使用高空间分辨率数据或只有高光谱分辨率数据进行分类的地图的准确度,分别为58.78%和77.54%。

Using climate model simulations to assess the current climate risk to maize production - IOPscience


  1. 对气候变率相关的收益冲击的研究相对较少。
  2. 短观测产量记录不足以对自然年际变率进行抽样,从而限制了概率评估的准确性。
  3. 美国和中国玉米种植地区提供了全球60%的玉米。
  4. 两个地区同时发生重大影响事件的可能性估计高达每10年6%,并且出现在物理可信的气候状态。
  5. 宏观尺度联合分析全球玉米产量