Rice plant counting, positioning and size estimating method based on DPS-Net deep learning

The invention discloses a DPS-Net deep learning-based rice plant counting, positioning and size estimation method. The method comprises the following steps of: inputting an original image of a rice field into a feature extractor, and extracting four feature maps with different scales; in a density estimation module, the attention map is fused with the initial density map based on a positive and negative loss function to generate a high-quality density map, and all pixel values of the high-quality density map are added to obtain the number of plants; in a plant position detection module, a non-maximum suppression algorithm is combined with the high-quality density map to generate coordinates of a plant position; in the plant size estimation module, the size of the plant is estimated by fusing the output of the module network structure with the plant position coordinates; a new high-throughput rice plant counting data set is utilized to prove that the method can realize automatic, non-contact and accurate count