Identification and Mapping of Individual Trees from Unmanned Aerial Vehicle Imagery Using an Object-Based Convolutional Neural Network

Precision agriculture (PA) is an agricultural management strategy founded on the observation, measurement, and response to inter/intra-field crop variability. It includes advances in data collection/analysis and management, along with the technological developments in data storage and retrieval, accurate positioning, yield monitoring, and remote sensing. The last provides unprecedented spatial, spectral, and temporal resolution, but can also provide detailed information about the vegetation’s height and various observations. Today, the success of new agricultural technologies has meant that many farming tasks have become automated, notably the identification and counting of trees individually. Scientists and companies have carried out more studies based on intelligent algorithms that automatically learn decision rules from data. The use of deep learning (DL) and particularly the development and application of some of its algorithms called convolutional neural networks (CNN) are considered a particular success. In the present work, we applied and tested the performance of an object-based convolutional neural network to automatically detect and map olive trees from a Phantom4 advanced drone imagery. The workflow involved image acquisition and the ortho-mosaic generation with Pix4D software, besides the use of geographical information systems and object-based image analysis. The application to an RGB ortho-mosaic of an olive grove, in the east region of Morocco, performed well achieving an overall accuracy of 97% and 99% after the OBIA refinement.