Investigando técnicas de processamento de imagens com IA na detecção de ferrugem em folhas de soja

Abstract

With the increase in world population, it is necessary to increase agricultural production. Since one of the problems faced to faced by farmers are pests and diseases in crops, which can cause larges damage to crops, the development of technologies that minimize losses is essential to increase productivity. Some of these technologies care for the environment, and among them stand out as they are built for the automatic and early detection of diseases through Unmanned Aerial Vehicles (UAVs) in conjunction with a computational method of deep learning. The literature in the area lacks comparative studies of such methods. The main objective of this study was to investigate the accuracy and efficiency of three computational methods, implemented in the form of image processing techniques and neural network architectures, without rust detection problems in soybean leaves. To map the methods applicable to the problem, we perform a systematic literature review. We characterize techniques for detecting diseases in plants and generate a set of important information that established an initial basis for research by selecting three methods for training Convolutional Neural Networks (CNNs): Original Imagens, SLIC, and Augmentor. The methods were tested and compared, generating quite motivating results that prove the efficiency of using image data augmentation techniques as an essential preprocessing step. In conclusion, the studies carried out point to the Augmentor method as the one with the best pre-processing results, being, therefore, the most interesting algorithm for training CNNs when applied in the early detection of soybean leaf rust. In addition, there is also a range of new research possibilities that can be applied to smart farming and precision agriculture. Key-words: Precision agriculture. Smart Agriculture. Machine Learning. Detection of Diseases. Deep Learning. Image processing. Convolutional Neural Networks.