Machine learning circuit design for processing neural networks in image sensor
Convolutional neural network (CNNs) have shown significant growth in the recent years as an effective algorithm to solve complex image recognition problems. Currently CNNs are being employed in a wide range of fields to tackle even higher number of problems which include face recognition, image clas...
Saved in:
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Thesis-Master by Coursework |
| Language: | English |
| Published: |
Nanyang Technological University
2020
|
| Subjects: | |
| Online Access: | https://hdl.handle.net/10356/141480 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Institution: | Nanyang Technological University |
| Language: | English |
| Summary: | Convolutional neural network (CNNs) have shown significant growth in the recent years as an effective algorithm to solve complex image recognition problems. Currently CNNs are being employed in a wide range of fields to tackle even higher number of problems which include face recognition, image classification and image segmentation. The past decade has witnessed an increasing need for continuous mobile vision which require image capturing and processing of vision features. Limitations largely due to intensive computation that requires expensive dedicated graphical processing unit has restricted further harnessing of the mobile vision technology. Moreover, the large analog input data required for processing increases the analog readout which has its own associated problems. This dissertation introduces the convsensor which is based on In-sensor computing technology. The convsensor has hardware pre-processing of input data which would effectively be reducing the complexity of the data to be processed in further stages of image classification and segmentation. |
|---|