Original papersCalculation method of surface shape feature of rice seed based on point cloud
Introduction
Rice is the most important cereal crop in China, and its planting area accounts for 30% of the total grain crop planting area in China. The purity of rice seeds is a key factor affecting the rice yield. The lower the purity of rice seeds, the lower the rice yield. Since rice seeds are often mixed with other varieties, coarse cereals and seed parents, finding a fast, lossless and accurate method of identifying rice seed purity is of great importance.
The surface shape feature of seeds is a very important index of classifying and recognizing seeds, and the morphological identification of seeds is a key method of recognizing seeds in biology (Kurtulmu and ünal, 2015, Szczypiński et al., 2015, Mebatsion et al., 2012, Mebatsion et al., 2013). In traditional seed identifying methods, manual identification is conducted according to the differences in surface morphology, which has tedious operations with large identification errors, and only coarse identification can be made. Currently, studies on rice seed purity detection based on seed surface morphology features mainly use computer vision technology to extract the shape, color, texture and moment features of seeds for seed classification and recognition. For example, (Wang et al., 2010) extracted 13 geometrical features of corn seeds, including area, circumference, long axis, short axis, etc., and 12 color features, including the mean value of each component of RGB, HSI color spaces, and applied neural networks (NN) for the classification of 4 species of corn seeds, with recognition rate larger than 97%. Zhu et al. (2012) used hyper spectral images of corn seeds to extract the entropy information of specified bands, and applied partial least squares projection algorithm to extract 65 optimal band features for corn seed classification and recognition, with recognition rate up to 99.19%. Zhao and Wang (2011) extracted the two-dimensional features of weed seeds. Nine invariant moments, including the ratio between seed perimeter and hilum perimeter, the ratio between seed area and hilum area, etc., as well as 7 invariant moments based on Hu’s moment were extracted. Weed seed classification was performed based on these features. Meizhi (2010) processed rice seed images and extracted texture features including entropy, energy, inertia moment, local stationarity, etc., and EQP texture feature distribution. Support vector machine was used to classify rice seeds. Shouche et al. (2001) extracted 6 shape features of wheat seeds, including area, perimeter, long axis, etc., and moment features, including original moment, invariant moment, central moment, normalized central moments, etc. The difference of shape features and moment features in representing wheat species was analyzed. Zapotoczny (2011) extracted 21 texture features in 7 color components for representing wheat seeds, and used them to classify 11 species of spring and winter wheat, achieving 100% classification accuracy. Choudhary et al. (2008) obtained 51 shape feature sets, 93 color feature sets, 56 texture feature sets and 135 wavelet feature sets by image processing. Then individual feature sets and combined feature sets were used for classification, and the recognition accuracy of feature set combination was 99.4%, 99.3%, 98.6% and 98.5% for wheat, rye, barley, and oats, respectively.
The aforementioned methods are simple and easy to implement, and the detection and recognition of different species can be achieved under certain conditions. However, the two-dimensional image features are the simplification of three-dimensional images of seeds, and the three-dimensional features of seeds are not considered, which limits the detection accuracy. Therefore, in both theory and real applications, the features of seeds should be extended to include the third dimension. In this study, we focus on rice seed purity identification problem, and propose to use the laser point clouds to extract the shape features of rice seed surfaces, then discuss the calculation method of the three dimensional parameters of rice seeds, which provides reliable three-dimensional data for later rice variety identification and other industrial application.
Section snippets
Experimental samples
The test materials are Yujing No. 6 rice seeds (Fig. 1). The seeds are sown in May, with
- •
Growth period: 150 days;
- •
Plant height: around 100 cm;
- •
Main stem leaves: 16 or 17;
- •
Spike length: 15–17 cm;
- •
Average grain number per spike: 110–130;
- •
Seed setting rate: 90%.
The grains have an oval shape, weigh 25–26 g per thousand grains, and have highest yield per unit area over 800 kg.
Three-dimensional laser scanning system
The adopted three-dimensional laser scanning system is composed of the control module, the driver module and the measurement module. The
Accuracy analysis of feature calculation
The time-consuming of collection and preprocessing of rice seed point clouds is about two hours if you can operate machine and software skillfully. A seed of Yujing No. 6 rice is randomly selected, and arbitrarily placed for three times. Its feature values are measured and calculated at each time, and the results are shown in Table 1. The error of the calculation results for the 3 times is less than 1%, indicating the algorithm is translation and rotation invariant.
Verification of the calculation method for length, width, thickness and volume
50 seeds of Yujing No. 6 rice
Conclusions
On the basis of the three-dimensional point cloud of rice seeds obtained by laser scanning, the surface shape feature parameters are obtained by point cloud model calculation. The errors are relatively small compared with the actual measured values.
- (1)
PCA method is used to perform translation and rotation transform for rice seed point cloud model. Then, the OBB of the point cloud model after transformation is calculated. The length, width and height of the OBB are used to represent the length,
Acknowledgements
This work was supported by the Nanjing Research Institute of Agricultural Mechanization Ministry of Agriculture [grant number: 201602004]; the National Natural Science Foundation of China [grant number: 51305182].
References (20)
A new clustering algorithm based on near neighbor influence
Expert Syst. Appl.
(2015)- et al.
Classification of cereal grains using wavelet, morphological, colour, and textural features of non-touching kernel images
Biosys. Eng.
(2008) - et al.
Probabilistic triangles for point set surfaces
Comput. Graphics
(2015) - et al.
Evaluation of variations in the shape of grain types using principal components analysis of the elliptic Fourier descriptors
Comput. Electron. Agric.
(2012) - et al.
Automatic classification of non-touching cereal grains in digital images using limited morphological and color features
Comput. Electron. Agric.
(2013) - et al.
Towards 3D Point cloud based object maps for household environments
Robot. Auton. Syst.
(2008) - et al.
Shape analysis of grains of Indian wheat varieties
Comput. Electron. Agric.
(2001) - et al.
Identifying barley varieties by computer vision
Comput. Electron. Agric.
(2015) Discrimination of wheat grain varieties using image analysis and neural networks. Part I. Single kernel texture
J. Cereal Sci.
(2011)- et al.
3D object retrieval with multi-feature collaboration and bipartite graph matching
Neurocomputing
(2016)
Cited by (36)
Accurate modelling and fracture characteristics of wheat grains
2024, Journal of Stored Products ResearchSeedscreener: A novel integrated wheat germplasm phenotyping platform based on NIR-feature detection and 3D-reconstruction
2023, Computers and Electronics in AgricultureDigital techniques and trends for seed phenotyping using optical sensors
2023, Journal of Advanced ResearchSurvey for smart farming technologies: Challenges and issues
2021, Computers and Electrical EngineeringAn Effective Denoising Method for the Point Cloud of Trees Based on the Hybrid Filtering Scheme
2023, Automatic Control and Computer Sciences