Employing Feature Extraction, PCA and Neural Networks to Identify Crop Blight for Precision Agriculture Applications

The domain of agriculture is being transformed through the use of UAVs and machine learning applications, often termed as precision agriculture. It is the amalgamation of data science, analytics, AI and ML technologies for enhancing conventional agricultural practices. This paper addresses the challenge of identifying crop blight  based on a machine learning approach. In this approach, the image is first pre-processed to convert from RGB to Grayscale and subsequently denoised. Next the statistical features of the image are computed followed by PCA for feature optimization. The optimized feature set is used to train a machine learning model with a penalty factor. The penalty factor is included acts as a regularization parameter to avoid vanishing gradients and overfitting through optimized weight updating mechanism.  The final classification accuracy is computed based on the TP, TN, FP and FN rates. Results show that the proposed model combining feature extraction, PCA and neural networks outperforms existing baseline models in the domain.