Combining Digital Image Processing and Machine Learning is Useful for the Early Detection of Salinity and Drought Stresses in Cucumber

Timely detection of plant abiotic stresses and their type and severity can be beneficial in order to prevent the loss of yield in crop production systems. This study introduces an image processing-based method to detect the type and severity of salinity and drought stress, as crucial abiotic stresses, in cucumber plants. Plants were cultivated in the greenhouse environment. Plant morphological features in the form of textural features were measured from the images captured from the plants. Sampling was performed five times at 3-day intervals beginning with applying the abiotic stresses. Measurements were conducted by transferring three leaves randomly selected from each pot to a chamber with artificial lighting equipped with a camera for image acquisition. The artificial neural network (ANN) was used based on the image textural features of the leaves as inputs and stress type and severity as output. The parameters of the network were optimized using a sophisticated optimization algorithm to achieve the most efficient machine. As a robust evolutionary method, the genetic algorithm was used to optimize the architecture of the ANNs. The results revealed that the image textural features for training ANNs optimized using the genetic algorithm could predict the type and severity of the plant abiotic stresses with MSE and R2 values of 0.092 and 0.74, respectively. Since the machine was able to perform a reliable stress prediction within a short period after applying the stress, the proposed method can be used for the early detection of salinity and drought stresses in cucumber plants.