Integrating IoT Sensors and Deep Learning for Robust Crop and Weed Discrimination in Dynamic Agricultural Environments
Anantha Sivaprakasam S, Senthil Pandi S, S Prathima, V Varshini
Abstract
In natural, unmanaged fields, telling the difference between crops and weeds is still very hard. This is a big problem for automated farming, especially when it comes to controlling weeds. There are many ways to get rid of weeds, but they are mostly tried and improved in controlled environments, which means they aren't very useful in real life. When used in real field settings, which are often more complicated and variable, these methods often don't work as well. To build a strong weed control system, you need to be able to tell the difference between the different types of weeds that are growing in the field and the crops that are being grown. This understanding is the basis for putting in place targeted and successful weed control plans that follow. Because natural field conditions are always changing and include many different types of crops and weeds, it is necessary to create complex and flexible systems that can reliably tell the difference between wanted crops and unwanted plants. Accurately achieving this discrimination is essential for the effective implementation of automated agricultural practices, which leads to more efficient and long-lasting weed control in real-life agricultural landscapes. In current research work, we have used an IOT sensor to collect real-time images of weeds and crops. Then images are pre-processed using image processing techniques such as image enhancement, resizing, etc. Then VGG-16 is used to perform the classification task. Finally, if the classified result is a weed, then with the help of a microcontroller spray the fertilizer using a motor have been activated which utilizes relay control. Also, the performance of deep learning architecture reports with better accuracy of 96.7%.