Grading of Pomegranate using Non Invasive Techniques

Abstract

newline Nowadays, the growth of quality control and agricultural monitoring system has newlineemerged a lot because of the increased requirement of sustainable food supply. Due to the newlinevarying climatic condition, the pomegranate fruits contaminate the neighbourhood fruits newlinewhen one is affected with diseased or insecticides. The pomegranate is one of the essential newlinefood source useful in providing high nutrition values and also effective for medicinal newlineapplications. Effective shelf life prediction approach is highly required to that must withstand newlineunder any climatic conditions. For the past few years, researchers are highly interested in newlinecomputer vision techniques for verifying the agricultural goods and predict the shelf life newlineeffectively. However, these techniques are suffered due to increased time consumption as newlinewell as sensitive to varying climatic conditions. newlineRecently, artificial intelligence (AI) based deep learning (DL) technique have newlinewitnessed the outstanding performance in predicting the shelf life under any climatic newlineconditions. This article brings the two novel objectives to predict the quality and shelf life of newlinethe pomegranate fruit accurately. newlineIn the first objective, three major processes are undertaken namely pre-processing, newlinefeature extraction and classification. At the pre-processing stage, the internal quality of the newlinepomegranate fruit is detected using magnetic resonance imaging (MRI). The scanned internal newlineportion is then transferred into feature extraction stage where the physiochemical features and newlinethe GLCM based features are extracted. Finally, LSTM (Long-Short Term Memory) model newlineand Stacked Dense Deep LSTM (SDD-LSTM) based DL model is proposed to predict the newlinefruit quality efficiently. Some of the effective qualitative analysis like square of correlation newlinecoefficient (r2), root mean square error of calibration (RMSEC) and root mean square error of newlineprediction (RMSEP) are analysed. newlineIn the second objective, three major steps are performed namely pre-processing, feature newlineextraction and classification. At the