Remote sensing detection and mapping of plastic greenhouses based on YOLOX+: A case study in Weifang, China

Plastic greenhouses (PGs) can provide suitable growth environments for crops, which has led to their extensive use in agricultural production. To estimate crop yields and prevent agricultural pollution, a strategy is required for the quick and accurate monitoring of PGs. However, due to their dense distribution, challenges remain for the precise identification and monitoring of PGs, as these data cannot be extracted simultaneously. In response to these issues, this study proposes an improved deep learning model based on YOLOX for the highly precise extraction of data pertaining to the quantities and areas of PGs. We developed accurate orientation boxes through the addition of angle detection heads and biased samplers. A dynamic balance strategy was introduced into the network to improve the learning capacity of the model. Meanwhile, the confidence of each slice was introduced into the NMS process to improve the reliability of the detection results. We compared our model with other object detection models (R3detnet, YOLOX, BBAvectors) of the greenhouse remote sensing image dataset, and finally employed the remote sensing images of Weifang City for the monitoring and mapping of PGs. The results revealed that: 1) Our improved model, which was dedicated to the simultaneous extraction of the quantities and areas of PGs, had the highest detection accuracy (average precision (Intersection over Union (IoU) = 0.5) = 88.19 %) in contrast to the other models; (2) The approximation of the IoU as a two-dimensional Gaussian distribution and calculation of the Wasserstein distance effectively assisted with the establishment of an oriented bounding box during detection; (3) The reliable extraction results of PGs in the study area confirmed the stability and validity of the model for practical applications; (4) The PGs in Weifang City were concentrated in the northwestern region, while their spatial distribution was affected by agricultural development patterns and geographical factors.