Data mining application in production management of crop

In this research, several image processing and data mining approaches were used to predict corn yield before harvest. A multi-temporal analysis was conducted to find the best date correlating to the actual yield. Corn aerial images of mid season (July 2nd half or August 1 st/2nd half) vegetation vigor were found highly correlated with the yield. ISODATA and neural SOM classification techniques were used to cluster the images based on the land-uses. Various vegetation indices, such as NDVI, GVI, SAVI, and PVI, were used to extract spectral features from the images. These features were used to develop crop yield estimation models. Bare soil spectral features (April 2nd or May 1st half); climatic data, such as temperature, accumulated growing degree days (AGDD), solar radiation, and rainfall; and non-climatic non-imagery data, such as irrigation and applied nitrogen, were also used as inputs to develop neural network prediction models. These prediction models could predict the corn yield around 2–3 months before harvest.; Two different neural network prediction-modeling techniques, back-propagation neural network (BPNN) and radial basis function network (RBFN), were evaluated and used for yield prediction model development. Several data transformation techniques, such as normalization, log₁₀, and inverse transformations, were evaluated, and the best one was used. Another study was conducted using the textural features of individual grids. The learning vector quantization (LVQ) classification technique was used to model the data to predict the yield in a range format. A color calibration technique was developed to diminish the radiometric interference at the time of aerial image acquisition. The predicted results were mapped into yield maps using the geo-statistics procedure.; Finally, another linear fit model was developed using the predicted yield from grid plots to estimate the seasonal water requirement of corn. The linear fit model for water use prediction from the predicted crop (corn) yield provided a correlation coefficient (r) of 0.72. However, residual analysis suggested the presence of outliers. The model created using the dataset after the omission of the outlier could enhance the model correlation status with r = 0.81. The model was compared with several other studies.