Predicting Wheat Yield Using Sequential and Deep Convolutional Neural Networks

Predicting Wheat Yield Using Sequential and Deep Convolutional Neural Networks

연구 분야: Artificial Intelligence

학회: 2024 5th International Conference on Electronics and Sustainable Communication Systems (ICESC)

Estimating yield and monitoring agricultural practices are crucial for ensuring local and global food security. Almost every region or country is somehow facing loss of yield due to change in climate and other conditions, it is essential to recognize the yearly variability in agricultural production and its link to meteorological conditions. If it is possible to forecast the accurate yield on time, it will surely help in the food maintenance as well as in the development of the country in the field of agriculture. Numerous factors are responsible for the crop of growth like quality of soil, temperature affect, pesticides used, rainfall measures and how the implantation works and manages on the agricultural land. In this study, we are using metrological parameters for prediction of yield with its accuracy. The effectiveness of Deep Convolutional Neural Networks (DCNN) and Sequential Neural Networks (SNN) for predicting wheat yield is investigated in this research. We use a strong dataset that includes yield records from the past, climate information, and multi-temporal satellite imagery. While DCNNs collect spatial characteristics from satellite imagery, sequential models-such as Long Short-Term Memory (LSTM) networks-capture temporal connections in climatic patterns. The Sequential model yields an R2 value of -0.11, while the Deep Convolutional Neural Network (DCNN) achieves a significantly improved R2 value of 0.216. The observed loss is 0.0260, with a mean absolute error also at 0.0260. The DCNN demonstrates an accuracy of 90.00% and an R2 value of 0.216, indicating its superior performance compared to the Sequential model.