Application of digital image processing in determining the physical properties of artificial orchid seeds by comparing different models of artificial neural networks and response surface methodology

Artificial seed production is a developing technology in plant biotechnology that provides considerable potential for conserving rare, endangered, and threatened species that are difficult to propagate through conventional propagation methods. In the present investigation, Image J software was used to process the digital image of the capsules and determine the volume, sphericity, and artificial seed centrality index. Response surface methodology (RSM) and artificial neural network (MLP and RBF) were utilized to model, predict and optimize the physical properties of the artificial seed of Phalaenopsis orchid. The training and validation data sets of the studied models were compared using regression coefficients (R) and sums of squared errors (SSE). Artificial seeds were obtained through encapsulation of Phalaenopsis orchid protocorm using various levels of sodium alginate concentration (3, 4, and 5%), calcium chloride concentration (100, 125, and 150%), and drop height (1, 1.5, and 2 cm) as input variables. Physical properties of artificial orchid seeds, including (capsule volume, sphericity index (SI), and concentric index (CI)) were considered as output variables. The results showed that image processing is an effective method for determining the physical properties of artificial seeds. According to the regression models, Results showed that interaction effects of the main factors are positively correlated with volume (p< 0.05). The effects of sodium alginate× calcium chloride concentration were significantly correlated with the CI index (P< 0.05). In addition, the sphericity value was positively correlated with sodium alginate concentration. In general, based on R2 and SSE values, the MLP model showed a much more accurate prediction than RSM and RBF models in terms of the values R2 and SSE. The optimum condition for volume, sphericity and CI index values was obtained using MLP 3-10-1 (R= 0.79, SSE= 0.0014), MLP 3-14-1 (R= 0.57, SSE= 0.0031) and MLP 3-14-1 (R=0.67, SSE= 0.0042), respectively. It is concluded that the ANN (MLP) is a desirable tool for the prediction and optimization of the physical properties of Phalaenopsis orchid synthetic seed.