Experimental investigation of crop-disease detection and crop-yield analysis systems: A numerical outlook.

Crop imagery is categorized into 3 different types, which are near-field images, satellite images and drone-based images. All these image types can be processed to determine crop growth, crop diseases and finally crop yield. Different algorithms have been proposed over the years which determine one or more of these parameters using a series of image segmentation, feature extraction, feature selection, classification and post-processing steps. Each of these steps requires a specialized set of algorithms to be employed to design an effective crop-image processing system. Due to the wide variety of algorithms present in the given field of work, the selection of the most optimum algorithm set for a given application is often ambiguous. For instance, if an application is trying to process satellite imagery, then identification of the best image-fusion methods for effective classification requires a lot of research, and thus increases the delay in designing the system. To reduce this ambiguity, this paper reviews these algorithm sets which identify the best techniques in terms of statistical parameters for a given application. Accuracy and error rates have been compared between different algorithms to give a clear idea about the performance of these algorithms. This comparison also allows researchers to identify best practices in terms of algorithm selection. [ABSTRACT FROM AUTHOR]