Skewness correction and quality evaluation of plug seedling images based on Canny operator and Hough transform

Abstract

The acquisition of quality information for plug seedlings is the foundation for automatic seedling transplanting. Machine vision technology is an extensively used method for this task. In the visual inspection of plug seedlings, the skewness of the images of dense-cell seedlings and the accuracy of target extraction affect the quality evaluation of plug seedlings. This study proposed a skewness correction algorithm on the basis of Canny operator and Hough transform for the images of a plug tray to improve the visual inspection method for transplanting equipment. Watershed algorithm was used to segment overlapping leaves, and gravity center method was applied to distinguish transboundary leaves. The leaf area and number of seedling leaves in the images of plug trays were extracted and used for quality evaluation. Two-week-old seedlings of Salvia splendens in 200-cell plug trays in a greenhouse were the objects of this study. Industrial cameras were used to capture images of the plug trays. These images underwent grayscale conversion in 2b–g–r channel, median filtering, Canny contour detection, and Hough transform to complete contour detection and skewness correction. The corrected angle deviation of the trays was less than 0.85°. Pre-treatment with grayscale conversion in 2g–r–b channel, binarization, and watershed algorithms were adopted to extract the target of the seedlings in the images. The image of the plug tray was divided into 200 small square areas in accordance with the cell area. The gravity center of the seedling leaves in the cell was calculated to locate the seedling. The number of the leaves and the area of seedlings in each cell were obtained and used as criteria to discriminate the quality of each seedling. The skewness angle of the plug tray during the actual delivery of plug seedlings was within the range of ±3°. Four sets of plug trays with skewness of ±1°, ±2°, and ±3° were obtained. Subsequently, the quality of seedlings in the four sets of plug trays was identified. The evaluation accuracy in the tray skewness and the angle correction situations were determined. Results showed that the average accuracy of seedling evaluation is 98% after angle correction. In contrast to the uncorrected images, the corrected ones can increase by 1.1–9.4 percentage points, thereby improving the evaluation accuracy for automatic seedling transplanting.