Drone technology and digital image analysis have enabled significant advances in precision agriculture, especially in site-specific treatment of weed escapes in crop fields. This study evaluated a pipeline for weed detection in multispectral drone imagery, along with site-specific herbicide application, using a remotely piloted aerial application system (RPAAS) targeting late-season weed escapes in rice with a selective postemergence rice herbicide, florpyrauxifen-benzyl. The efficacy of the RPAAS-based herbicide application with geocoordinates of weed escapes obtained manually or based on image analysis was compared with conventional backpack broadcast spray. The weed species targeted were barnyardgrass, Amazon sprangletop, yellow nutsedge, and hemp sesbania. A Python-based rice–weed detection model was developed using the canopy height model and spectral reflectance of weeds and rice plants. Results indicate that the accuracy of image-based detection for late-season weed escapes in rice was highest for hemp sesbania (95%), followed by Amazon sprangletop (87%) and yellow nutsedge (74%), with barnyardgrass showing the lowest accuracy at 62%. The study found that the backpack broadcast method had the highest efficacy in weed control, followed by the RPAAS method using manually obtained geocoordinates and those based on image analysis. Site-specific herbicide application using RPAAS resulted in a 45% reduction in herbicide compared to the broadcast backpack application. Moreover, the RPAAS site-specific application method for late-season treatment minimized the field area affected by herbicide injury and protected rice grain yields compared to the broadcast method. Overall, the utility of unmanned aerial sprayer–based detection and site-specific treatment of late-season weed escapes in rice has been demonstrated in this research, but further improvements in weed detection efficacy and the accuracy of targeting plants with RPAAS are necessary.