Advancing Crop Disease Identification Using Machine Learning and Transfer Learning Approaches

Abstract

Convolutional Neural Networks (CNNs) have changed how we handle image processing. Different types of CNNs offer various benefits in how well they perform and how efficient they are with computing resources. This paper looks at five well-known CNN types: ResNet151, Xception, DenseNet201, InceptionV3, and EfficientNetB7. Each has special features that help improve deep learning. ResNet151 uses skip connections to solve the problem of vanishing gradients, allowing very deep networks to be trained for recognizing images and detecting objects. Xception builds on the Inception design by using depthwise separable convolutions, which makes it more efficient while still performing well. DenseNet201 connects layers closely, promoting better flow of gradients and reuse of features, which helps in tasks that need efficient computing. InceptionV3 includes multi-scale convolutional layers to optimize computing costs while maintaining high accuracy, making it great for large image classification. Finally, EfficientNetB7 uses a scaling method that achieves top accuracy with fewer parameters, making it effective for tasks that require precision and efficiency. The paper compares these CNN types and their uses, showing their influence on computer vision.