A DIGITAL IMAGING MODEL FOR PLANT STRESS DETECTION

Abstract

Detection of stress in plants has been one of the many difficulties faced by the farming industry in the developing world. This difficulty has been attributed to a variety of factors, namely; reduced food production and consequent food insecurity, and limited access to computing technologies that result to technological evolution challenges among Kenyan farmers. With respect to the cause of these difficulties, research indicates the need to experiment with a diversity of image processing techniques, and formulation of algorithmic models that would tackle the challenge in plant stress. These stresses require technological advances in image processing and algorithms that can be used by local farmers in detection of plant stresses. Digital image processing approaches based on relevant algorithms allows for precision farming through detection and remedying of foreseen stress before it causes destruction on the plants in the farmers’ fields. To achieve this, the general objective of this study was to develop a digital imaging model for detection of plant stress for enhanced productivity of crops and food security. In order to realize the general objective, the study was guided by the following specific objectives; to analyze existing image-based plant stress detection approaches; to establish the physical features of stress in plants; to map the physical features into digital imaging signature characterizing stress in plants; to develop a digital imaging model for plant stress detection and to validate the digital imaging model for detection of plant stress. This study was guided by positivism research philosophy, and explored existing models, algorithms and image processing techniques for detection of plant stress, and studied the growth of mobile telephony with relation to the need for food security. An experimental research design was employed through embracing the Convolution Neural Networks. The study resulted to a model that was developed and implemented in a mobile application and web interface to enhance food security through detection and monitoring of tomato pests and disease stress. The images were captured using mobile phone cameras, acquired input images were preprocessed through resizing and rescaling of images, whereas Gaussian blur, thresholding and dilating were used in feature extraction. SoftMax was used for classification and optimization carried out using the Adam Optimizer. Validation for accuracy of the model was based on training steps, sets and epochs and TensorBoard was used in bid to validate the model. The results of the study proved the reliability of the model in detection of stress using the digital imaging model to be more efficient over traditional approaches of plant stress detection. The findings of this study if adopted will contribute to increasing food production and enhance food security, contribute to the body of knowledge on digital imaging technology, and inform academic researchers, computer scientists, policy makers in education, and all stakeholders in farming and agriculture, on the implementation of new and emerging deep leaning technologies in improving existing work.