Palm Trees Counting in Remote Sensing Imagery Using Regression Convolutional Neural Network

Date palm trees are important economic crops in many countries and counting their numbers in a plantation area is crucial information for predicting the yield of date fruits, determination of insurance and financial aids, etc. In this abstract, a supervised tree counting framework is proposed using Convolutional Neural Network (CNN). The proposed approach casts the counting process into a regression problem, instead of following the classification or detection framework. To further decrease the prediction error of counting, we fine-tuned a pretrained CNN architecture into regression model. As the final output, not only the tree count is estimated for an image, but also its spatial density map is provided. Trained with small image patches cropped from airborne dataset, the proposed method is compared to manual counting and obtains good performance.