A Two-Stage Low-Altitude Remote Sensing Papaver Somniferum Image Detection System Based on YOLOv5s+DenseNet121

Papaver somniferum (opium poppy) is not only a source of raw material for the production medical narcotic analgesics but also major certain psychotropic drugs. Therefore, it stipulated by law that cultivation must be authorized government under stringent supervision. In areas, unauthorized and illicit on private-owned lands occurs from time to time. These illegal sites are dispersedly-distributed highly-concealed, therefore becoming tough problem The low-altitude inspection unmanned aerial vehicles has advantages high efficiency saving, large amount image data collected needs manually screened, which consumes lot manpower resources easily causes omissions. response above problems, this paper proposed two-stage (target detection classification) method sites. first stage, YOLOv5s algorithm was used detect images purpose identifying all suspicious original data. second DenseNet121 network classify results so as exclude targets other than retain containing only. For achieved best overall performance among mainstream target models, with Precision 97.7%, Recall 94.9%, mAP 97.4%. pre-training performance, classification accuracy 97.33% 95.81%. experimental comparison between one-stage suggest two methods remained same, reduced number falsely detected 73.88%, greatly reduces workload subsequent manual screening remote sensing images. achievement provides an effective technical means solve in supervision cultivation.