Problem Statement 1: A thing of one innovative idea where ML-AL-DS can be used, describe it in 100 words.
India shares 3,323 km of its border with Pakistan. Often some parts of the border are left unprotected due to the sheer length of the border. Intruders can easily infiltrate the border and enter the country. The army cannot place personnel at every inch of the border. Perhaps this problem can be solved with the help of Deep Learning. Night vision cameras module enabled with GPS can be placed along the border. These cameras are connected to a cloud system which hosts a neural network pre-trained to recognize humans. When a camera recognizes a human, the GPS co-ordinates of that camera are sent to the command centre. From this the personnel are alerted to check if there is some activity going on at that location. The border areas are now equipped with optic fibres, so a strong internet connection is available for communication in real-time. The feed of real-time video can be accessed at the centre. That way the soldiers can be better equipped to deal with the situation. These cameras can operate throughout the day thus easing the jobs of soldiers.
What’s the best life hack you have invented recently?
-
As a foodie I often find myself browsing through Instagram and looking at delicious dishes but unable to find the name of the dish to order it for myself. So I trained a ResNet model to classify images of food. I used the food-101 dataset for this. Presently the network can identify 10 types of food that include burritos, pies and so on. With the help of a powerful machine (and more patience) it can be trained to recognize over a 100 types of dishes. The beauty of this network is its very specific. For example it doesn’t just recognize a pie as just a pie but also the type of pie such as apple pie. This can be further improved to post the photos of food images and write a caption for the same on social media. It can also be used to order food directly. The future scope of this project would be take an image of the food and an app can automatically find that item and order it for you thanks to the captioning from the neural network.
-
My final year project was a wearable health device that takes the ECG of a person and inform him if his heart was healthy or if he suffered from a condition called ‘atrial fibrillation’. Atrial Fibrillation (AF) is caused when the rhythm of the beating of the heart goes haywire. This condition can cause strokes and render the heart weak making it prone to heart attacks. Our device extracted ECG from the person and fed to a neural network which was trained on 8,500 odd ECG signals and tested on 4,000 odd signals. The output of the network classified the ECG as either ‘Normal’ or ‘AF’. The monitoring happens in real time. Since AF is highly episodic the person can wear this device and go about doing his daily chores. As and when he is alerted by the device he can seek further medical help.
Please open Cough1.ipynb jupyter notebook for the solution for Problem Statement 2.