Mid to Late Season Weed Detection in Soybean Production Fields Using Unmanned Aerial Vehicle and Machine Learning

Mid-late season weeds are those that escape the early season herbicide applications and those that emerge late in the season. They might not affect the crop yield, but if uncontrolled, will produce a large number of seeds causing problems in the subsequent years. In this study, high-resolution aerial imagery of mid-season weeds in soybean fields was captured using an unmanned aerial vehicle (UAV) and the performance of two different automated weed detection approaches – patch-based classification and object detection was studied for site-specific weed management. For the patch-based classification approach, several conventional machine learning models on Haralick texture features were compared with the Mobilenet v2 based convolutional neural network (CNN) model for their classification performance. The results showed that the CNN model had the best classification performance for individual patches. Two different image slicing approaches – patches with and without overlap were tested, and it was found that slicing with overlap leads to improved weed detection but with higher inference time. For the object detection approach, two models with different network architectures, namely Faster RCNN and SSD were evaluated and compared. It was found that Faster RCNN had better overall weed detection performance than the SSD with similar inference time. Also, it was found that Faster RCNN had better detection performance and shorter inference time compared to the patch-based CNN with overlapping image slicing. The influence of spatial resolution on weed detection accuracy was investigated by simulating the UAV imagery captured at different altitudes. It was found that Faster RCNN achieves similar performance at a lower spatial resolution. The inference time of Faster RCNN was evaluated using a regular laptop. The results showed the potential of on-farm near real-time weed detection in soybean production fields by capturing UAV imagery with lesser overlap and processing them with a pre-trained deep learning model, such as Faster RCNN, in regular laptops and mobile devices. Advisor: Yeyin Sh