Smart Weed Management and their Profile Mapping using Deep Learning and Digital Mapping

Removal of weeds has been a challenging task for farmers. Conventional method of weed removal includes indiscriminate spraying of herbicides despite the fact that the presence of the weed is patchy. To find a better solution for weed management, two prototypes were developed using low-cost hardware and open-source Deep learning models: Prototype-1, used a Raspberry Pi 4 B and YOLO v5 model while Prototype-2, used a more powerful NVIDIA GTX-1080 Ti GPU and YOLO v7 model for weed detection. Both prototypes were evaluated in custom potato fields, targeting common weeds like Chenopodium album and Portulaca oleracea while avoiding potato plants. The prototype-1, due to its low processing capabilities showed a significant lag in real-time detection of weed while the prototype-2 showed promising results with average precision of 97% and recall of 91%. A GPS device and Digital Mapping system was also integrated to calculate the effectiveness of Herbicide after evaluating the before and after results with the help of Smart Agro-Sprayer. These experiments show a promising future for the fusion of latest technologies like AI (Deep learning), Automation, Digital Mapping and Data analytics to solve conventional agriculture problems and promote precision farming to increase crop yield with sustainable model.