Deep Learning - Segmentation, Identification and Classification
summary
Uses multispectral drone imagery from a Phantom4 Pro to map soil chemistry and estimate tree age by training convolutional nets, then segments objects with selective search and crops proposals into a network for classification and pixel-level segmentation. Implements models with Caffe + DIGITS, transfers ImageNet weights, extracts feature vectors and compares with Euclidean/Hamming distances. Reports GPU speedups (10,000 iterations ~50 minutes on CPU vs under 3 minutes on GPU, about 5x faster) and shows the pipeline runs end-to-end.
AI-written summary · about
chapters (10)
- 0:00Introduction and background
- 6:13Machine learning basics
- 11:10Deep learning applications
- 15:40Deep learning frameworks
- 21:06Convolutional neural networks
- 24:04GPU vs CPU performance
- 27:54Using Caffe framework
- 31:39Image classification demo
- 33:28Object segmentation methods
- 35:38Resources and conclusion
related talks
18mDeep learning for Space imagery applications
36mClassification of Satellite Imagery
48m(Deep) Learning to Fly
46mDeep Learning for Satellite Image Processing, Lessons Learned
33mWhat and where? - Machine learning for geospatial image analysis
45m(Deep) Learning to Fly
11mUsing AI to Accelerate the Smart Green Transition
26mDetecting and classifying object images using ruby