Skip to main content
ARS Home » Northeast Area » Beltsville, Maryland (BARC) » Beltsville Agricultural Research Center » Environmental Microbial & Food Safety Laboratory » Research » Publications at this Location » Publication #379597

Research Project: Sensing Technologies for the Detection and Characterization of Microbial, Chemical, and Biological Contaminants in Foods

Location: Environmental Microbial & Food Safety Laboratory

Title: Classification of watermelon seeds using morphological patterns of X-ray imaging: A comparison of conventional machine learning and deep learning

item AHMED, MOHAMMED - Chungnam National University
item YASMIN, JANNET - Chungnam National University
item PARK, EUNSUNG - Chungnam National University
item KIM, GEONWOO - Orise Fellow
item Kim, Moon
item WAKHOLI, COLLINS - Chungnam National University
item MO, CHANGYEUN - Kangwon National University
item CHO, BYOUNG-KWAN - Chungnam National University

Submitted to: Sensors
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 11/25/2020
Publication Date: 11/26/2020
Citation: Ahmed, M.R., Yasmin, J., Park, E., Kim, G., Kim, M.S., Wakholi, C., Mo, C., Cho, B. 2020. Classification of watermelon seeds using morphological patterns of X-ray imaging: A comparison of conventional machine learning and deep learning. Sensors. 23(20), 6753.

Interpretive Summary: Because seed viability is an important seed quality factor affecting seedling health and production yields, there is a great need for nondestructive methods to accurately determine seed viability prior to seed sale and planting, particularly for fruit and vegetable crops. This study analyzed X-ray projection images of three varieties of watermelon seeds to find morphological features useful for classification of the seeds into two categories: normal viable, and abnormal nonviable. These features were then used to test several conventional methods of classification as well as several methods based on convolution neural networks (CNN). Results showed that convolutional neural network methods could outperform conventional classification methods. The best accuracy from the latter group was 83.3% from conventional linear discriminant analysis, while the best CNN accuracy was 87.3% from a 50-layer deep residual network (ResNet-50), demonstrating that the transfer learning approach of the CNN methods combined with X-ray projection imaging is feasible for classifying seeds based on morphology.

Technical Abstract: A seed’s morphological quality plays an important role in germination. A seed with a good internal structure can supply sufficient nutrients during germination and produce healthy seedlings. To investigate the internal condition of seeds, X-ray projection imaging was adopted to capture images of three varieties of watermelon seed. Several image processing techniques, such as intensity and contrast enhancement, and noise reduction were applied to the images. Then, various types of image features (i.e., local binary pattern, Gabor, local Fast Fourier Transform (FFT), texture, contrast and Haralick textural (Tx) features) were extracted and three classifiers (linear discriminant analysis (LDA), quadratic discriminant analysis (QDA), and k-nearest neighbors algorithm (KNN)) were assessed using the sequential forward selection (SFS) method and Fisher objective function for selecting features in conventional classification. For transfer learning approaches, simple ConvNet, AlexNet, VGG-19, ResNet-50, and ResNet-101 were used to train the dataset and class prediction. A germination test was carried out to identify two seed classes: (1) normal viable seeds and (2) nonviable and abnormal viable seeds. In the conventional classification, LDA performed the best, with an 83.6% accuracy using 48 features. ResNet-50 performed better than other transfer learning architectures, with an 87.3% accuracy which was the highest accuracy in all classification models with 10-fold cross-validation. The findings of this study demonstrate that transfer learning combined with X-ray projection imaging is a feasible approach for classifying seeds based on their morphology.