I have wanted to write more to redevelop the habit of writing regularly. While I didn't get time to write a blog post yesterday, I will spend a bit of time doing it now by pondering my areas of interest over the next few weeks. I delayed the job search a few weeks to give myself more unstructured time to play around with machine learning projects. In actual fact, I expect much of the next few weeks will be spent on specifically honing my skills to get a job rather than wandering through the intellectual wilderness. Still, there are a few job types dealing with analyzing aerial images and computer vision for robots that I would like to feel more confident in applying to.
 |
| Here is a picture that I thought was a cute illustration of computer vision for robots from IronOx - ironox.com - which is working on robotic farming. Computer vision for robots is not always so literal, with the cameras looking like eyes. |
Aerial imagery and vision for robots seem like the most compelling applications of computer vision I have come across, and I think generally gaining a skillset in computer vision is something I would like to acquire. There are a few specific areas that I would like to pursue projects. I already have a project going that seeks to detect oil well's from satellite images. (For anyone curious, the project is at
https://github.com/dzubke/oil-well-detection/ and click the .ipynb file with the most recent date with the YYYYMMDD suffix.)
I will focus primarily on this project in the immediate future and depending on how successful my algorithms are at performing this detection, I will incorporate some elements of change detection into this project. "Change detection" involves analyzing one image and then analyzing another image of the same thing at a future time and detecting certain changes between the two images. I have heard that change detection has been used to track the progress of oil well drilling, so that would be an obvious extension of my existing project. For the record, I am not wildly interested in oil wells. They just happen to surround my hometown and I have looked at many satellite images of them and, thus, it was an easy place to start. And oil well detection seemed more interesting (although perhaps more challenging) than detecting swimming pools.
 |
| Oil wells as seen from Google Maps imagery. |
 |
| Versus some swimming pools. I'm now to so sure swimming pools would be easier, though their high levels of blue compared to the surroundings seems like an easy manner of detection. Though after just 30 seconds of my scanning for them, tennis courts and the odd blue rooftop would be likely false positives for a swimming pool detector. |
I addition to focusing on computer vision, I want to generally improve on my programming skills. In some cases, data science and software engineering are considered separate skillset with some obvious overlap. However, in certain places like Google, you need to pass a certain benchmark of software engineering skills for any data science position. I suspect this minimum threshold exists at more place than I think. After having many friends or their partners going to med school, I can appreciate that there is a corpus of knowledge necessary to enter a certain field, and so I am respectful of (and somewhat excited by) the need to acquire this basic level of software engineering literacy. My copy of "Cracking the coding interview" is already on its way. I'll take a few courses in more advanced topics in Python.
Thanks for reading. I'm not exactly sure what the "tone" of my future blog posts will be, although I may just write whatever I feel like, whenever I feel like. That seems a good guiding principle for now.
