Comparison of image-based methods for determining the inline mixing uniformity of pesticides in direct nozzle injection systems

Seven image-based methods to determine mixing uniformity in inline mixers by image processing were compared using viscous pesticides mixtures with gradually increasing uniformity. After obtaining initial optimal methods and calculation parameters, they were further evaluated by a jet mixer and a layered mixer. Results showed that quantitative descriptions for uniformity inside the tube by grey level co-occurrence matrix (GLCM) and concentration difference within time sequence (CVT) failed to reflect accurately the changing-trends in uniformity. A histogram information entropy (HIE)-based method tended to underestimate changes, while coefficient of variance based on each single frame (CVS) which calculated differences in pixel values underestimated it when mixtures showed stratification, indicating that both methods were not suitable. However, the accuracy of methods using explicit features such as second-order moments of grey-scale histogram (HSM) and optimised area-weighted uniformity index (OAU), as well as implicit methods such as principal components analysis (PCA) were suitable, and the normalised index > 80% can be suitable because no non-scattered pesticide remained. Single-viewing images were adequate for evaluating the uniformity produced by mixers although small errors between horizontal and vertical-viewing images existed. During evaluations by mixers, OAU and PCA proved to be more advantageous than HSM because these methods do not only rely on pixel probability distributions of grey-scale histograms, but the positional information of the pesticide distribution, even though the HSM is more time-efficient than them.