Long short‐term memory based real‐time monitoring of potato slice drying using image chromatic features

Abstract

Listen

Translate article

AbstractThe computer vision system provides a better alternative to conventional quality assessment methods during drying because it is rapid, inexpensive, nondestructive, sensitive, and precise. Experiments were conducted for drying of 1‐mm‐thick potato slices at 45, 50, 55, and 60°C. A laboratory‐scale hot air dryer with the provision of installing an image acquisition system was used to visualize the chromatic changes during the drying process. Multiple input single output (MISO) type long short‐term memory (LSTM) network was used to study the time series data. This network has multivariate input data (chromatic attributes of potato slices) to predict single output (moisture content). The optimized network after hyperparameter tuning has an architecture of 86 hidden layers, an LSTM depth of 2, a dropout value of 0.4, and a learning rate of 0.01, providing the highest R of 0.96 and the lowest RMSE of 13.25 × 10−2. During performance evaluation, the R and RMSE values were obtained to be 0.98 and 3.75 × 10−2, 0.97 and 4.29 × 10−2, and 0.98 and 2.67 × 10−2 at 47, 54, and 58°C, respectively.Novelty impact statementThe real‐time monitoring system helps minimize human intervention to assess product quality during drying. This system allows us to comprehend the complexities in the chromatic quality changes in the product while drying. It also finds its usefulness in continuous type dryers where quality monitoring is a serious issue. Furthermore, the same algorithm can be used for multiple operations like cooking, roasting, baking, etc., and various agricultural commodities after adequate training.