Design of A High Efficiency Augmented Machine Learning Model For Multiparametric Crop Type And Yield Prediction

Abstract

Smart farming is a multidisciplinary process that involves predicting the type and quality of
plant that would give maximum yield under a given set of geographical conditions. This
involves taking into account factors such as climate, soil, and other environmental factors.
The vast majority of intelligent farming systems accomplish this by employing deep learning
models, but these models have poor performance in terms of accuracy, scalability, and the
ability to deploy in real time. This is because the majority of deep learning models require
large amounts of training data, which in turn increases the amount of time that must pass
before they can be deployed. In addition, the applicability of a model that has been trained on
one particular type of crop to other types of crop is limited. This article proposes an enhanced
machine learning model for multiparametric crop type and yield prediction as a means of
enhancing the functionality and applicability of deep learning models. The 'you look only
once' (YoLo) model is fused with the combination of the VGGNet-19 model, the Inception
Net model, the Xception Net model, and the GoogleNet model that the proposed model uses.
This allows for high-speed processing. The model makes use of multiparametric information,
such as plant imagery, soil parameters, weather information, temporal geographical data, and
nutrient information, in order to predict the type of crop that can be grown under a given set
of environmental conditions and the maximum yield that can be obtained from that crop.
When the proposed model is evaluated against a diverse range of crop and soil types, it is
found to possess a high degree of accuracy, achieving a score of 98.7% in terms of crop-type
prediction and 97.6% in terms of yield prediction. When the proposed model is analyzed in
comparison to other models that are considered to be state-of-the-art, the results reveal a
performance improvement of 8% in terms of average accuracy, 6% in terms of average
precision, 3% in terms of recall, and 6.5% in terms of area under the curve (AUC). The
suggested model also demonstrates a delay reduction of 9%, which qualifies it for use in realtime
high-speed deployments. This text also highlights a number of different case studies that
were conducted in order to validate the performance of the model, and it also recommends
some improvement areas that can be further researched in order to achieve an improved
application-specific performance levels.