Intelligent crop yield prediction system using neural networks and databases

Abstract

Machine learning plays an important role in decision support and yield forecasting. This is an effective tool for determining strategies during the growing season of plants. The article proposes a method for predicting yield using a complex system consisting of a convolutional neural network (CNN), a feedforward neural network (FNN), and a SQLiteStudio database. The system includes several stages of data processing, starting with the collection and analysis of images and digital data obtained from various sources, and ending with yield forecasting based on this data. A convolutional neural network (CNN) is used to analyze images and video streams to recognize and count fruits on trees, providing accurate data about the status of the crop. Feedforward neural network (FNN) is used to analyze digital data, such as weather station data and long-term crop yield data, to subsequently predict crop yields. The received data is stored in a relational database, which ensures their structured storage and access for subsequent processing. Used SQL language to perform various database operations. To automate the process of counting fruits on trees, the YOLOv8 convolutional neural network model is used, which allows recognizing objects in real time. A Python script has been developed to process images using YOLOv8 and save the results to a database. An integrated system combines various methods and technologies to predict yields and automate data collection and analysis processes. The developed model showed a mean square error (MSE) of 7.33 and a mean absolute percentage error (MAPE) of 6.27%.