Summer internship at Krsikx
One of the key experiences recommended during your time as an under graduation is doing an internship. gaining the work experience is a key for a good employability. for that I decided to go for the company where I can learn and enhance my skill. startups are the best place to learn a new ideas and get the amazing experience of corporate world. fortunately, I got selected in a AgriTech startup-Krsikx india as a intern.
I undertook 5 week internship as a ML/DL developer. I was one with the 33 university level student interns. some of them were my fellow adamic interns. unfortunately, as there was a pandemic situation ongoing, we have to do a remote internship. however, it was fun and we met together every evenings at online meetings and talk about difficulties and solution.
Project and work
At the start of the week we were given a task to find a problems in agriculture and to provide the solution for that with the help of technology.
There are mainly 3 projects I had worked on but because of the problem in dataset we completed 2 out of 3. though it was a great achievement.
project 1- crop water requirements
it’s regression problem in which we need to find the requirements of the water of the crop. its a very hard thing to find the trend or features from data as its very noisy. data cleaning and pre-processing techniques also not showing result as good as want. so we halt that project and start working on the next.
project 2 — weed detection
Main aim for the project is to classify the crop and weed in real time using IOT and ML.
for that we have a dataset contains of weed and crop images. also the labeled data of the crop and weed was given in .txt format.
Task is to train the object detection model on the dataset and make the improvements for real-time application.
dataset contains the 1300 images of weed and crops includes with labels.
firstly, we analyze the data and divide into train and test set. import all the files in the Drive and also configure the Darknet. advantage of the google colab is that its provide a free GPU for ml-trainings. model was traind and saved in drive. although, traing took 4+ hours to train.
I have used a YOLOv3 and YOLOv4 object detection to train. we get a good result with YOLOv4.
here, is the result of our model.
project 3- PDD
PDD - Plant Disease detection. there are many disease of crop that we can find by just looking at the leaf of the crop. the project aims to find the disease based on the leaf. all interns were allocated a 2–3 crops and dataset. for me its a Rice and Potato. I have used mobilenet and ResNet for training respectively.
Here is the result of accuracy for potato and rice.
conclusion:
to sum it all, it was an amazing experience to work with a startup. I have created a 2project based on my skill and knowledge and provide best to the company. however, we faced many difficulties with data but problems are the first step towards the success.
keep learning, keep experiencing..!
