Plant Disease Management using AI/ML techniques

This is the project I worked on as a Junior Research Fellow at Dhirubhai Ambani Institute of Information and Communication Technology. The team had to build an early prediction system for fungal disease detection. To do that, we built in-house soil moisture and leaf wetness sensors. These were deployed in the field and used to monitor the plant growth conditions. Environmental parameters like ambient temperature, ambient humidity, and leaf wetness duration were recorded.

I assisted in the data collection process and further built algorithms to process the collected sensor data. To gather useful information for the predictive models, I built pipelines for data pre-processing and feature extraction. I honed my research skills and performed an extensive literature review to find the best applicable methods for sensor data processing. The data was cleaned using mathematical algorithms like Total Variation Denoising for denoising and baseline correction of data. This was imperative to extract accurate information from the sensors. Since the code was implemented in Python, I gained valuable experience and honed my skills.

For building predictive models, we relied on using neural networks and hence, I learned how to build complex neural networks from scratch. I worked on Tensorflow and PyTorch to build neural networks. A notable project was building an "attention-based multi-input multi-output neural network for plant disease prediction". Attention is a mechanism popularly used in transformers in the domain of NLP. Through these mechanisms, transformers are trained to pay "attention" to the more relevant elements of the input. I applied this method to our data with the intuition that the model would give a higher weight to the more relevant inputs while training to enhance the prediction accuracy.

 IoT-based Pantry Management System for Households

For our final year project, we came up with the idea of developing a pantry management system for households. The inspiration behind the project was to ease user experience by providing them with an up-to-date information about all the kitchen ingredients available at their houses. The project execution was two-fold.

First, we had to develop a sensitive analog front-end which was essentially a data acquisition system. It was composed of load sensors to measure the weight of the drawer, RFID sensors to identify which ingredient was added or removed and various other transducers to transmit the data to the micro-controller. Sensor integration was applied to optimize the sensor network. Arduino-Uno micro-controller was used to control and process the data collected.

The next part was to integrate the system with a cloud network so that all the data could be accessed by the user from a remote location through an application. NodeMCU was the firmware used to handle the IoT operations by creating a gateway.