Apple Trees Leaves Pathologies Detection using Deep Learning Convolutional Neural Network

Effective agricultural governance is key in countries that rely on this sector of the economy for secure and reliable development. The appearance of plant pathologies and their classification plays a crucial role. Diagnosing diseases using traditional approaches based on expert assessments is not accurate enough, takes a long time to process data, and contains subjectivism in the assessments. In smart agriculture for digital image processing, machine learning is a powerful tool for automatic detection of plant diseases. Utilization of deep learning and convolutional neural networks in particular allows more precise image analysis and more accurate disease detection. In this paper an approach for detecting the pathologies in apple trees’ leaves using DenseNet-based architecture is presented. To account for the interference on cameras operating in real natural environment a “white noise” synthesized by a chaos-based pseudo-random number generator is added to the image data. The accuracy achieved after training the model on a dataset comprising 1800 images in 4 categories (healthy, scab, rust, and multiple diseases) is 99.91% for the training data and 98.25% for the test data, which proves the reliability and accuracy of the model for detecting potential plant infections.