We present an automatic agave detection method for counting plants based on aerial data from a UAV (Unmanned Aerial Vehicle). Our objective is to autonomously count the number of agave plants in an area to aid management of the yield. An orthomosaic is obtained from agave plantations, which is then used to create a database. This database is in turn used to train a Convolutional Neural Network (CNN). The proposed method is based on computer image processing, and the CNN increases the detection performance of the approach. The main contribution of the present paper is to propose a method for agave plant detection with a high level of precision. In order to test the proposed method in a real agave plantation, we develop a UAV platform, which is equipped with several sensors to reach accurate counting. Therefore, our prototype can safely track a desired path to detect and count agave plants. For comparison purposes, we perform the same application using a simpler algorithm. The result shows that our proposed algorithm has better performance reaching an $F_1$ score of 0.96 as opposed to 0.57 for the Haar algorithm. The obtained experimental results suggest that the proposed algorithm is robust and has considerable potential to help farmers manage agave agroecosystems.

Building inspection searching for superficial defects, such as cracks, is a vital task because such damages cause economic losses or put at risk the integrity of people. For this reason, different ways to reduce the costs and risks through the use of robotic systems that allow make inspections have been studied. Among these robotic systems, we have the unmanned aerial vehicles (UAV) that allow reaching difficult access places permitting better inspection. In this work, we propose using convolutional neuronal networks for crack recognition from images captured by an UAV. To carry out the training task of the network, a database of cracks in walls was built from images collected from the Internet. The training of the network prompted encouraging results with a 95% accuracy over the training set. Experimental results of crack recognition in images were carried out validating the application of the proposal.