Pragmatic Analysis of Fruit Processing Models for Quality Evaluation from a Statistical Perspective

Fruits & other naturally occurring plant-based products change their temporal visual appearance & smell when separated from their parent trees. The temporal visual appearance is used by image processing analysts to evaluate its quality, and lifespan related parameters. In order to perform this task, various components are used, which include, pre-processing models, segmentation models, feature extraction & selection techniques, classification models, and post-processing techniques. Each of these micro models vary in terms of performance metrics like accuracy of detection, error rate, signal to noise ratio (SNR), computational complexity, delay of processing, cost of deployment, etc. Due to such a wide variation in performance, there is an ambiguity in method selection while designing an efficient fruit quality detection model. Due to this ambiguity, researchers need to validate individual model performance, and then interface them with their own architecture. This process is cumbersome, and increases deployment cost & time needed for bringing the product into market. In order to reduce this ambiguity, a review of recently proposed models for fruit quality detection is discussed in this text. These models are reviewed on the basis of their nuances, advantages, limitations, and future research scopes along with their scalability performance. This text also compares the reviewed fruit-image models in terms of processing delay, accuracy, precision, recall, computational complexity, and recommended application of use. These applications include quality analysis, artificial ripening detection, disease detection, and fruit grading applications. Based on this comparison, researchers can identify best performing model(s) and/or their combinations for application-specific deployments.