Article Preview
Top1. Introduction
Deep learning (DL) and flutter are the high-tech solutions in computer science. Deep learning methods have significant strides in visual identification applications. Detecting and categorizing image datasets with deep learning is progressive research (Abdel-Hamid et al., 2014; Alzu’bi et al., 2020; Zhang et al., 2019). DL offers a solution to accomplish artificial intelligence enabling machines to solve problems in such a way as the human brain does. It has been implemented in several areas of people’s day-to-day lives. For example, Siri and Alexa, the voice-based apps are utilized to recognize the speech to help and enable people to communicate with devices by speaking instead of keyboard based inputs.
An ordinary work in image processing is recognizing the same kinds of items using machine learning techniques for classifying and clustering the fruits (Bhole et al, 2020b). Previously, a thermal imaging idea in Naik and Patel (2017), machine learning with a color feature extraction model (Bhole et al, 2020a) and deep neural network (Bhole et al, 2020c) have been suggested and manifested for evaluation of quality and categorizing maturity of the mango (Behera et al., 2020; Sahu and Potdar, 2017; Sultana et al., 2017; Bhole and Kumar, 2020d). Numerous techniques have been employed for identifying the quality of fruits with maturity levels through RGB imaging, either under-controlled or real situations (Behera et al., 2020; Sahu and Potdar, 2017; Sultana et al., 2017; Intaravanne et al, 2012). Particularly, these non-destructive methods have been dependent on the examination of visible and spectral imaging using pattern analysis. Moreover, very little work has been performed based on thermal imaging for fruit quality evaluation. The benefit of the thermal imaging approach is that it is contactless, non-destructive, faster, and also works well in dark environs to lighting environments as well as helpful in addressing the social problems and creating the applications seamlessly. It offers an opportunity for creating new ways to inspect processes that have not been even possible previously.
Furthermore, the number of published works, particularly on the evaluation of shelf-life of fruits, is yet restricted to particular cases. Fan and Zhou (2011) utilized Near-Infrared Spectroscopy to inspect the trait of apple at different periods of shelf life and obtained the accuracy up till 100%. But, the use of Near Infrared Spectroscopy is expensive and not handy technique. However, this is a destructive approach which was used at the laboratory level and can’t be useful for real-time evaluation. Zhang et al. (2013) employed VIS/NIR spectrometer to assess the different stages of Malus Asiatica Nakai fruit by observing the variation in pectin of the fruit. However, this is a destructive approach which was used at the laboratory level and can’t be useful for real-time evaluation. Nandan Thor (2017) predicted the shell life of banana by extracting color features along with 7 machine learning methods and claimed the maximum accuracy as 52%. This approach is non-destructive but the features are extracted manually through image processing techniques by using external surface parameters with very little accuracy.
The automatic shelf-life prediction system has become essential, mainly in the agriculture and food business to reduce the human-intensive work processes and this can be done on mobile devices (Karar et al., 2021) to enhance efficiency. But, this would need sophisticated feature identification process to evaluate fruit shelf-life using similar features that have a high color resemblance. Therefore, customized machine learning models are required to evaluate fruit’s shelf-life and address the intricate features (Bhole et al., 2021) of images for particular applications.