Desarrollo de una aplicación para la detección de plagas en el cultivo de papas aplicando redes neuronales convolucionales

Abstract

In recent years, the development of intelligent applications has intensified considerably, making it easier to solve complex tasks or tasks that require great effort. Different investigations show that artificial intelligence plays a very important role in different areas such as medicine, robotics, education, agriculture, mathematics, etc., which, thanks to the advantages offered by this type of technology, has allowed find more efficient and effective solutions to the needs of the population, by automating processes. Recent studies carried out on the use of Artificial Intelligence in agriculture, establish that its use has allowed to improve and innovate traditional systems of cultivation, irrigation, soil identification, weed identification, where the benefits are reflected in the increase in production of quality crops. Agriculture is one of the oldest areas of study in humanity and therefore it is considered a fundamental pillar both in the feeding of thousands of people and in the internal and external economy of a country, therefore, producing products of Quality is an important need for farmers. The potato is one of the products with the highest demand in many countries and is one of the most cultivated crops at the national level, both on small and large scales. Commonly, potato cultivation is affected by different factors that cause the appearance of pathogenic agents, the same ones that cause severe diseases. That is why identifying pests in potato crops in time is key to avoiding economic losses. The fastest and safest way to identify these pests is through the use of artificial intelligence technology. In order to support farmers to have a better healthy management of the crop, it is intended to develop an application, using artificial intelligence tools for the identification of pests with great precision in the potato crop through the use of CNN convolutional neural networks. To carry out this project, the Jupyter Notebook tool was used for the design of the convolutional neural network, Visual Studio Code and PyCharm as source code editors and libraries such as TensorFlow, Uvicorn, FastApi among others. During the construction of the neural network, it was decided to divide the data, 80% of the data for training, 10% for validation and the remaining 10% was used for the evaluation of the model, in order to obtain satisfactory results. For the design of the application in this case, a web application was chosen to use React Js, which fits perfectly to our needs, since it allows us to implement user interfaces in a simple and fast way. For the validation of the implemented model, multiple tests were carried out and by using the confusion matrix it was possible to calculate the performance metrics of the prototype, obtaining an average accuracy of 90%. Thanks to the results obtained from the application with real data, it is concluded that our model presents an acceptable and reliable level of precision, with a precision higher than the average necessary to consider the model valid.