Weed Detection in Cotton Production Systems Using Novel YOLOv7-X Object Detector

A weed is a wild, undesired plant that grows naturally along with the desired crop. For their growth, they compete with the main crop for various resources like space, sunlight, irrigation, nutrients, etc. This leads to an enormous loss in the yield of the main crop and hence needs to be selectively controlled. Human intervention in the process of identification of weeds and subsequently its removal is extremely tedious and time-consuming too. Achieving the desired level of accuracy and preciseness in the manual weed identification process is impracticable. In recent years, researchers have proposed computer vision-based methods for automatic weed identification in precision agriculture. In this paper, we have used YOLOv7-X for automatically detecting weeds in cotton production systems. YOLOv7-X is a relatively new addition to the You Only Look Once (YOLO) family of fast and accurate algorithms. The benchmark dataset used for the purpose of validating our results is CottonWeedid-15. This dataset is customized with annotations suitable for YOLOv7-X by using the roboflow tool. The experimental study demonstrates that the YOLOv7-X model’s mean average precision (mAP@.5) can attain 96.6 %. The average precision and average recall of the model were 0.914 and 0.953 respectively. This model can also be used to classify several weeds in various crops.