Projects

The Sortobot project is aimed at the development of indigenous technology for sorting and grading of agro-commodities at an industrial scale. It uses computer vision to analyze the surface of instances of the commodities and uses machine learning for classification based on various user-selectable parameters such as the severity of the defect. With this project, we have developed a sorting machine that costs a fraction of the imported alternatives while at the same time has better accuracy, functionality, and flexibility that modern trade demands. Please visit Occipital Technologies in order to enquire about this product.

The Agrograde project brings the power of AI in the hands of marginalized farmers and small traders who are particularity vulnerable to consignment rejection and the resulting financial losses due to disputes over quality parameters of the vegetables and fruits. This app enables farmers to snap an image of randomly sampled instances of produce for quality analysis and grading. The computer vision algorithm does the state of the art instance segmentation and classifies each instance on the basis of size, color, and defect. Please visit Occipital Technologies in order to enquire about this product. I have also attached a demo video.

This project is aimed at the use of CNNs in Bio-Medical images for object detection purposes. Proteins are “the doers” in the human cell, executing many functions that together enable life. Historically, classification of proteins has been limited to single patterns in one or a few cell types, but in order to fully understand the complexity of the human cell, models must classify mixed patterns across a range of different human cells. Images visualizing proteins in cells are commonly used for biomedical research, and these cells could hold the key for the next breakthrough in medicine. However, thanks to advances in high-throughput microscopy, these images are generated at a far greater pace than what can be manually evaluated. Therefore, the need is greater than ever for automating biomedical image analysis to accelerate the understanding of human cells and disease.

In this project, I used custom CNN along with transfer learning to detect iceberg from satellite observations. I used 15 fold image augmentation to test the effects of data augmenation on satellite data. The score without image augmentation was 0.299 (2556th rank). After augmentation the score was 0.1571(400th rank). Couple of tweaks and optimazation may yet be needed to generate better result.

Since convolutional neural networks are great for images recognition, I modified the task of speech recogniton into image recognition by creating spectrographs from .wav file. Then I trained a custom CNN model and it worked decently, achiving 85% accuracy.

This project was aimed at exploration into the application of autoencoder-decoder based architectures in nerve segmentation. The dataset is sourced from kaggle. I trained UNets for semantic segmentation of nerves in ultrasound tiff images.