Assessment of Frogeye Leaf Spot Severity in Soybeans Using 3D Imaging Techniques

Visual assessments of the severity of plant diseases often vary among evaluators. A pipeline was developed for quantifying frogeye leaf spot, caused by Cercospora sojina, in soybean using three-dimensional (3D) imaging techniques. Soybean leaves were collected from field and humidity chamber environments following inoculation with C. sojina. Flatbed scanners were used to acquire conventional (2D) images, and a Peel 3D scanner was used to generate 3D models, producing JPG and OBJ/PLY files. Image segmentation was conducted using k-means clustering to classify 2D and 3D pixels into healthy, diseased, or mixed categories. Point cloud data from 3D scans were processed and analyzed with CloudCompare and R software to evaluate the percentage of leaf area occupied by lesions. Statistical comparison between 2D and 3D measurements produced a concordance correlation coefficient of 0.828, indicating substantial agreement, although the 3D method tended to overestimate disease severity at low levels (<10%) and underestimate severity at higher levels (≥10%). Lighting inconsistencies during 3D scanning affected accuracy of the estimates. Despite these limitations, 3D imaging shows potential as a complementary tool to traditional images for evaluating disease symptoms. Additionally, the dataset made available provides a foundation for future applications in machine learning and automated disease quantification in plant pathology. Continued refinement of imaging and analysis methods is essential for improving the accuracy of severity estimates and enabling high-throughput, objective phenotyping in disease quantification for comparison of treatments or screening for host resistance.