A non-destructive coconut fruit and seed traits extraction method based on Micro-CT and deeplabV3+ model

Abstract

With the completion of the coconut gene map and the gradual improvement of related molecular biology tools, molecular marker-assisted breeding of coconut has become the next focus of coconut breeding, and accurate coconut phenotypic traits measurement will provide technical support for screening and identifying the correspondence between genotype and phenotype. A Micro-CT system was developed to measure coconut fruits and seeds automatically and nondestructively to acquire the 3D model and phenotyping traits. A deeplabv3+ model with an Xception backbone was used to segment the sectional image of coconut fruits and seeds automatically. Compared with the structural-light system measurement, the mean absolute percentage error of the fruit volume and surface area measurements by the Micro-CT system was 1.87% and 2.24%, respectively, and the squares of the correlation coefficients were 0.977 and 0.964, respectively. In addition, compared with the manual measurements, the mean absolute percentage error of the automatic copra weight and total biomass measurements was 8.85% and 25.19%, respectively, and the adjusted squares of the correlation coefficients were 0.922 and 0.721, respectively. The Micro-CT system can nondestructively obtain up to 21 agronomic traits and 57 digital traits precisely.

Keywords: Micro-CT; coconut phenotypic traits; deep learning; non-destructive; plant phenomics.

Figure 1

Principal components and configuration of Micro-CT system. (A) Hardware design of the Micro-CT system. (B) The field of view for coconut fruit imaging. The active area of the panel detector is 248mm⨯298mm with a resolution of 2512⨯3008 pixels. The distance between the X-ray source and the detector (DSD) is 1200 mm, the distance between the X-ray source and the rotation center (Dsc) is 1050 mm; thus, the field of view is 217mm⨯260mm, and the spatial resolution can be calculated as 86μm⨯86μm. (C) The field of view for coconut seed imaging. The distance between the X-ray source and the rotation center (DSC) is 600 mm; thus, the field of view is 124mm⨯149mm, and the spatial resolution can be calculated as 49μm⨯49μm.

Figure 2

Image analysis pipeline of Micro-CT system. (A) As the coconut sample rotated, 360 X-ray projected images at different angles were acquired and calibrated; (B) A FDK algorithm was applied to obtain a reconstructed section image of the coconut fruit; (C) A DeepLabV3+ model was used to segment the slice image and acquire the area of the mesocarp, endocarp, solid and liquid albumen and cavity; (D) A local least squares and a Poisson surface reconstruction algorithm were introduced to convert the boundary points into a 3D model; (E) Filled the 3D model to obtain the 3D cloud point with CT value; (F) Using the segmentation results of the slice images, the 3D model was segmented.

Figure 3

The model structure of DeepLabV3+ with the Xception backbone.

Figure 4

flow chart of traditional image processing segmentation algorithm. (A) The slice image at the middle height of the sample. (B) The background area is removed by the Otsu algorithm and connected component algorithm. (C) The edge enhancement algorithm determines the coconut coat and inner shell area. (D) The seed points and threshold tolerance of solid and liquid albumen are manually set to obtain the segmentation result. (E) The slice image at another height close to (A). (F) Take the segmentation result (D) as the template to segment (E). (G) The slice image at another height far from (A). (H) Take the segmentation result (D) as the template to segment (G). A priori knowledge detects the cavity area at the center.

Figure 5

The coconut fruit projection images from the Micro-CT system, the corresponding 3D model from the structured light system, and the reconstructed slice images with some out-of-ranged samples. (A) An out-of-ranged immature sample with full liquid albumen. (B) A normal immature sample with full liquid albumen. (C) A normal immature sample with limited liquid albumen. (D) An out-of-ranged mature sample with limited liquid albumen. (E) A normal mature sample with full liquid albumen.

Figure 6

The coconut fruit projection images from the Micro-CT system and the reconstructed slice images. (A) A fast-growing sample with well-developed haustorium and sufficient albumen. (B) A sample that failed to germinate. (C) A normal sample with a developing haustorium and sufficient albumen. (D) A slow-growing sample with insufficient albumen.

Figure 7

The accuracy analysis of DeepLabV3+ model. (A) Sample with cavities and solid albumen. (B) Sample without cavities and have thin solid albumen. (C) Sample with thick solid albumen. (D) Sample with thick solid albumen and cavities.

Figure 8

The 3D point cloud model of coconut fruit and seeds, and the sectional views in three directions. (A) Result for a coconut fruit sample. (B) Result for a coconut seed sample.

Figure 9

Mesurement of (A) coconut fruit total volume, (B) total surface area, (C) milk volume, (D) copra weight, and (E) biomass.