Optimal Feature Extraction Technique for Crop Classification Using Aerial Imagery

Identification and classification of crops cultivated areas using satellite images is important for making different types of analysis to formulate policies and plan in agricultural and environmental domains. A range of Feature Extraction Techniques (such as Statistical, Texture, DWT and DCT), having a vital role in crop classification, have been employed and analyzed to suggest the best one for classifying crop images from regions of Netherlands and Pakistan. These feature extraction techniques have been evaluated with number of classification techniques like Support Vector Machine, Naïve Bayesian, K Nearest Neighbor, Decision Tree and ensemble based classifiers. A rigorous analysis has been carried out to evaluate the effectiveness of different machine learning techniques over variety of feature sets in the regions under study. Data set containing satellite images from cultivated lands of Netherlands and Pakistan have been used for experimentation. DWT and DCT have shown better precision and accuracy in classifying the crop images from Netherlands and Pakistan respectively, compared to rest of the feature extraction techniques.