Visão computacional aplicada a imagens do biospeckle laser em sementes

Abstract

In techniques used to analyze seed viability and vigor through the biospeckle laser phenomenon, there is a possibility of automating the image analysis process by means of computer vision techniques. This study aimed to develop a method for segmentation and detection of the spatial orientation of soybean seeds illuminated by the biospeckle laser. A total of 229 images of soybean seeds obtained from previous biospeckle laser experiments were used. The images were processed in the ImageJ® software, in which a macro was developed to automate the entire analysis workflow. Initially, a Gaussian smoothing filter was applied to reduce image noise. Then, different thresholding methods were tested, selecting the one with the best performance in image binarization. After binarization, a region-growing segmentation algorithm was applied. To determine the spatial orientation, the Hough Transform was employed to fit the ellipse that best represented the segmented seed shape. Before applying the method to real images, a test with objects of known geometry—rotated at different angles—was performed to verify the precision and robustness of the Hough Transform in detecting and fitting ellipses. The results confirmed the technique’s ability to correctly identify the spatial orientation of the objects, validating its application to biospeckle laser images of seeds. Finally, the angles calculated by the developed method were compared with real measurements obtained using a digital goniometer, allowing the calculation of the mean error. A sensitivity test to seed rotation was also performed. The method presented a mean error of 5.65% and a resolution of 0.0026°, demonstrating high precision and robustness in detecting the spatial orientation of the seeds. It is concluded that the developed technique is viable and reliable for application in biospeckle laser experiments with soybean seeds, constituting a promising tool for automated analyses. Keywords: computer vision; Hough transform; biospeckle laser; Glycine max.