A deep learning-based pin precision weeding machine with densely placed needle nozzles

Abstract

With advancements in artificial intelligence and robotic technology, the demand for innovative weed control methods has increased. This paper proposes a novel weeding machine concept that integrates artificial intelligence with micro-needle nozzles, enabling precise and selective herbicide application based on weed size and type. The system employs deep learning-based semantic segmentation to accurately identify weeds at the pixel level. Following identification, densely arranged needle nozzles deliver fine streams of herbicide directly to targeted weeds. The herbicide dosage is regulated by time-controlled shooting, facilitated by solenoid valves. The developed pin-precision weeding machine features a 1.20-meter-wide nozzle plate bar equipped with 128 injection needles, enabling simultaneous herbicide application to multiple weeds. In an open bean field, the detection accuracy of the proposed Spray-Net achieved a mean Intersection over Union (mIoU) of 88.6% for bean instances and 90.9% for weed instances. Furthermore, the system demonstrated a detection speed of 28 frames per second (fps) and a hitting accuracy of 86.1%. Notably, the proposed weeding machine boasts a weeding capacity of up to 4266 weeds per second with 128 nozzles in operation. The proposed pin-precision weeding machine represents a pioneering approach in environmentally friendly, intelligent weed management.