Identification of corn plants using aerial images obtained by UVAs

Precision Agriculture is a branch of agronomy that focuses on the use of precision technology in crop management. It is utilized for the management of cultivation processes and their georeferencing with the aim of improving the overall productivity of the plantation. This allows farmers to monitor their production more accurately and consistently. This study is justified because by determining the number of plants in a crop, one obtains information that directly influences the overall productivity of a plantation. This information can be used by the farmer in various ways, such as mitigating losses if the results are extremely detrimental to the crop or, in less severe cases, identifying external factors that may have affected the productivity of the planting. This enables monitoring of such conditions to reduce productivity losses in future plantings, as well as providing a financial estimate of the final production value.This work presents a method for identifying and counting corn plants, which involves using machine learning to train a neural network to identify corn plants. The neural network chosen for the experiments was RetinaNet, which was trained using a dataset generated from aerial images obtained by a UAV. This corn plant image dataset contains 3,129 labeled examples used for training and validation, and 757 image examples in the test dataset. The results obtained in the precision metric calculation showed good values compared to the literature, where the analysis was conducted using the test dataset, achieving a precision of 99.0% and a recall of 97.6%.