A Novel Approach for Crop Yield Prediction based on Hybrid Deep Learning Approach

Abstract

The agricultural sector is crucial to the economic development of our country. Civilization's birth was facilitated by agricultural practices. The agricultural sector is vital to India's economy because of the country's status as an agrarian nation. So, agriculture has the potential to serve as the economic foundation of our nation. In agriculture planning, crop selection is crucial. Our Indian economy desperately needs widespread reforms in the agricultural sector. In this proposed approach to use several machine learning methods to forecast future agricultural yields. After receiving the input image, the null values can be filtered out using the preprocessing approach. The Relief method is then used to choose features. In order to extract features, a linear discriminant analysis approach is used. Finally, the CNN-BiLSTM-ECA model, which combines a CNN, a Bidirectional Long Short-Term Memory network, and an Attention Mechanism, is presented for use in training (AM). To reduce the impact of excessive noise and nonlinearity, CNN has been used to extract deep aspects of agricultural productivity. Crop yield is predicted using a BiLSTM network trained on the recovered deep characteristics. This proposed also implement an unique Efficient Channel Attention (ECA) module to increase the network model's sensitivity to key features and inputs. The average error made by each method is compared to one another. Farmers will be able to use the CNN-BiLSTM-ECA forecast to guide their planting decisions by taking into account variables like expected temperatures, precipitation, available land, and more.