Development of an image-based android application for quality inference of tomato

Abstract

Abstract

The objective of this study is to develop real-time quality evaluation tool of tomato from image input. Recent advancements in deep learning tools such as Tensorflow Lite, have assisted in building a light weight real-time android-based application for tomato quality inference.  Availability of smart phones and its developmental prospects can meet the growing concern of consumers for quality foods from image. Deep learning has significant potential on image identification and hence an image-based application is thus opted. This work is an effort to develop an image-based AI tool for quality inference of tomatoes. To execute the task of application development, an extensive study on the quality attributes of tomato is done and different state-of-the-art CNN models are trained on tomato images for quality prediction. The proposed CNN models after being trained on tomato image dataset are then deployed in an android application for the following quality inferences: (a) prediction of current state of tomatoes as edible or spoilt, immature or partially mature, or fully mature (b) prediction of physico-chemical properties and (c) shelf-life estimation. Experimental results indicate high classification accuracy of 99% and 97% respectively for spoilage detection and maturity detection respectively from tomato images. In addition to the high recognition rate, the tflite models in android application consumes very less computation time and is able to make prediction in real-time (<0.67 sec). Thus, this application can be considered as a viable solution in tomato quality inference.

Keywords: Tomato, quality, app development