Search

A novel methodology combining microscopy observation with Artificial Neural Networks (ANNs) and realized by machine learning algorithms for the study of starch gelatinization was developed. As the most critical part during object detection, an improved starch single shot multi-box detector (starch-SSD) originated from ANNs was purposely designed and applied in monitoring the morphological changes of starch with increasing temperature. In the case, the birefringences were automatically identified by computer vision and then the relative birefringence number of the image was calculated. Basing on such number change, the temperature of phase transition was detected and consequently the degree of gelatinization (DG) at specific temperature was quickly calculated. Compared with traditional methods that mainly performed by manual operation, experimental results confirmed that the proposed method has competitive accuracy and is much faster. It also provides a unified standard for microscopy observation without subjective uncertainty.

A novel image segmentation methodology combined with optical microscopy observation was developed for qualifying starch swelling. Starch granules in the micrograph were successfully segmented based on high-precision edges extraction achieved by Canny edge detection together with mathematical morphology operation. Granules were automatically identified by computer vision and characterized by giving quantifiable area of these granules. The evolved swelling process could be generally divided into two phases. During the first phase, starch granules were only swollen up by 2.56 %, which is hard to be identified by conventional naked eye. During the following narrow temperature interval (60-66 ℃), these starch granules were detected to swell up significantly by 9.08 %. Through the granule area variable, swelling capacity was high-throughput characterized, which allows for the whole evaluation to be completed within a couple of minutes. The proposed methodology showed a high accuracy and potential as a novel technique for characterizing gelatinization.

A gelatinization degree control system, with a combination of Artificial Neural Networks (ANNs) and computer vision, was successfully developed. An intelligent measurement framework was purposely designed to achieve a precise investigation on phase transition and morphology change of starch in real time, as well as a process control during gelatinization. Base on a variation of birefringence number, the degree of gelatinization (DG) control system provided a direct and fast methodology without subjective uncertainty in studying starch gelatinization. In the course, the whole system was a cascade structure with the hot-stage temperature chosen as the inner-loop parameter, thus the granule morphology and birefringence at different DG could be easily observed and compared in real time, and the relative transition temperature was simultaneously calculated.