Machine Learning-based Estimation of Soybean Growth

Abstract

The ability to determine the yield of soybean crop and the extent of the area of interest with non-destructive methods and at minimal cost is very important because of the increasing world population and the effects of climate change on crop productivity. Yield mapping is important in the crop field at the initial stage of crop management since it gives the total yield in the field, and the distribution of yield in the field which help in the decision making process of which area to fertilize, to irrigate, or to spray pesticide. Thus, the goal of this study was to improve the yield of the soybean crop and the return on investment while utilizing the yield prediction with image data with the least amount of resources and environmental pollution. The Problem Statement is conventional procedures, which are mostly based on the observation and statistical approaches, are often complex, time consuming and prone to errors. That is why the achievement of accurate and timely decisions for crop management is impossible with such restrictions The development of the presented modern technologies implies the search for new approaches that would improve the assessment of the growth with higher accuracy. The proposed research study makes use of Convolutional Neural Network (CNN) to predict the soybean crop yield from an RGB image dataset. The significant phases include data acquisition phase, data pre-processing phase, model-building phase, model-assessment phase, and model-testing phase. The layers in the CNN architecture include the convolution layer, pooling layer, activation layer and the dense layer as well as the output layer. The training of the model was carried in three steps and before each step, the hyperparameters of the model was adjusted. The model is proved to perform well and reliable for yield estimation. The accuracy of testing of the model, is 92.50% and validation accuracy is 94.59% whereas, the training accuracy is 100%.