查看文章

The real-time detection of parthenium weed plants in the pulse crop field was carried out using low altitude flying drone. A fully convolutional semantic segmentation model was proved to accurately perform object segmentation with higher time complexity. In this research, the LinkNet model with Resnet-34 was used for real-time detection of weed plants using a video feed from low altitude flying drones. Experimental results is proven that LinkNet-34 can detect overlapping and irregular shape weed objects at 0.86 mean pixel accuracy of 0.598 mean IoU at 0.217 s. The processing speed was better compared to LinkNet and U-Net models. The detected weed images were stitched together to create a weed site map. The created map is automatically uploaded to google cloud for further site analysis.