Machine-vision based handheld embedded system to extract quality parameters of citrus cultivars

This manuscript introduces a handheld machine vision based system design that is capable of standalone operation using touch screen based user interface and also can operate through smartphone based android app. System uses 8.0 Megapixel, 1080p CMOS camera interfaced with quad-core ARM Cortex-A53 processor based computing platform (Raspberry Pi computing platform) for real time image acquisition and processing. Multi-spectral led array has been used to compensate the effect of external illumination and also to increase the accuracy of measurement. System stores acquired images on interfaced 16.0 G.B. external memory card with date and time information. Various segmentation methods have been explored to extract region of interest in acquired images and compared based on the capability of segmentation in real-time. Segmented images have been used to extract different features such as color, shape, size and texture using various image processing algorithms. Extracted features have been fused together and undergone through different statistical and neural network based modelling methods to correlate features dataset generated using handheld system with standard quality parameters of collected citrus samples. Performance of the established correlation models for various quality parameters such as chlorophyll, sugar content, TSS, weight, pH and volume have been evaluated and best performed models for each quality parameter has been used to train the developed handheld machine vision based system. Overall system is battery operated and also enables cloud connectivity using on-board Wi-Fi facility or smartphone based android app. Overall device has dimensions of 12.0 x 6.0 x 4.0 (in cm), weighs 139.07 g and runs with 5-V rechargeable battery.