Tiller number, an important growth parameter for rice cultivation, is still being assessed manually. This work investigated the influence of dataset composition on performance of deep learning models for tiller number estimation in rice. Four datasets were constructed for early tillering, active tillering, and maximum tillering by applying the concepts of mixed varieties, class balance, and data augmentation. YOLOv4 models were trained to estimate tiller numbers using each constructed dataset. Then their performance was evaluated. Results demonstrated that the models trained with datasets created using a combination of mixed variety, class balance, and augmentation showed the best performance for estimating the tiller number at the three tillering stages with a mAP range of 68.8–86.4 %.

View full abstract

Due to the fact that the airflow field cannot solve the accumulations of mixtures on the vibrating screen surface, a conical fan was designed. The CFdesign software was used to numerically simulate the airflow field of the cleaning room. The results showed that compared with the cylindrical fan, the mass of threshed mixture in the main falling area of the vibrating screen surface under the conical fan accounted for 65.4 %. The accumulation central point of the threshed mixture deviated 165 mm from the vibrating screen width to the grass discharge outlet. Under the conical fan, the impurity content of the mixture not screened was 6.8 %, and the cleaning loss rate was 0.2 %.

View full abstract

The production of sugar is adversely affected by unhealthy sugarcane, which decreases the yield and quality and is difficult to detect through non-destructive tests. This study aims to accurately differentiate between healthy and unhealthy sugarcane in a laboratory environment using a portable visible near-infrared spectrometer with multivariate analyses. The spectra of 100 each of healthy and unhealthy sugarcane parts are analyzed in this study. The classification rates of the partial least-squares discriminant analysis and support vector machine classification of healthy models are 100 % and 91.9 %, respectively, while the classification rates of unhealthy models are 65.2 % and 78.3 %, respectively. The overall results demonstrate that NIR spectroscopy and multivariate analyses are effective at classifying healthy and unhealthy sugarcane.

View full abstract