RESUMEN
Deep Learning models have presented promising results when applied to Agriculture 4.0. Among other applications, these models can be used in disease detection and fruit counting. Deep Learning models usually have many layers in the architecture and millions of parameters. This aspect hinders the use of Deep Learning on mobile devices as they require a large amount of processing power for inference. In addition, the lack of high-quality Internet connectivity in the field impedes the usage of cloud computing, pushing the processing towards edge devices. This work describes the proposal of an edge AI application to detect and map diseases in citrus orchards. The proposed system has low computational demand, enabling the use of low-footprint models for both detection and classification tasks. We initially compared AI algorithms to detect fruits on trees. Specifically, we analyzed and compared YOLO and Faster R-CNN. Then, we studied lean AI models to perform the classification task. In this context, we tested and compared the performance of MobileNetV2, EfficientNetV2-B0, and NASNet-Mobile. In the detection task, YOLO and Faster R-CNN had similar AI performance metrics, but YOLO was significantly faster. In the image classification task, MobileNetMobileV2 and EfficientNetV2-B0 obtained an accuracy of 100%, while NASNet-Mobile had a 98% performance. As for the timing performance, MobileNetV2 and EfficientNetV2-B0 were the best candidates, while NASNet-Mobile was significantly worse. Furthermore, MobileNetV2 had a 10% better performance than EfficientNetV2-B0. Finally, we provide a method to evaluate the results from these algorithms towards describing the disease spread using statistical parametric models and a genetic algorithm to perform the parameters' regression. With these results, we validated the proposed pipeline, enabling the usage of adequate AI models to develop a mobile edge AI solution.
Asunto(s)
Agricultura , Citrus , Algoritmos , Benchmarking , Inteligencia ArtificialRESUMEN
There is a constant risk of iron ore collapsing during its transfer between processing stages in beneficiation plants. Existing instrumentation is not only expensive but also complex and challenging to maintain. In this research, we propose using edge artificial intelligence for early detection of landslide risk based on images of iron ore transported on conveyor belts. During this work, we defined the device edge and the deep neural network model. Then, we built a prototype will to collect images that will be used for training the model. This model will be compressed for use in the device edge. This same prototype will be used for field tests of the model under operational conditions. In building the prototype, a real-time clock was used to ensure the synchronization of image records with the plant's process information, ensuring the correct classification of images by the process specialist. The results obtained in the field tests of the prototype with an accuracy of 91% and a recall of 96% indicate the feasibility of using deep learning at the edge to detect the type of iron ore and prevent its risk of avalanche.