Classification of Citrus Leaf Diseases Using Convolutional Neural Networks Compared to Traditional Neural Networks

India is an agrarian country with a diverse agricultural sector, covering approximately 157 million hectares (387.9 million acres) of agricultural land. It is considered the second largest in the world after the United States. The main objective of this research is to compare different algorithms for segmenting disease lesions, collecting shape-related features, and classifying these features for detecting disease spots (Pseudocercospora angolensis) and determining the severity of the disease. In k-means clustering, the value of k-means signifies the total number of clusters, and each cluster centroid defines the center of the region. Centroids are initially selected at various locations, and each point in the dataset is connected to the nearest centroid using the squared Euclidean distance algorithm. Centroids are then rearranged according to the average of their data points. This process is repeated until the centroids no longer change position, thus finding the optimal positions for the centroids. The k-means clustering segmentation technique divides an image into k-means clusters, grouping areas with similar properties together and separating them from other regions of the image. In machine learning, pattern recognition, and image processing, feature extraction starts from an initial set of measured data and builds derived values (features) that are informative and non-redundant. This facilitates subsequent learning and generalization steps and can lead to better human interpretations. Humans can better understand a reduced feature set compared to the whole image. Artificial Neural Networks (ANNs) are models that resemble biological neuronal structures. The starting point for any ANN is an elementary neural element that mimics the behaviour of real neurons. Each computing unit in an ANN is based on the concept of an idealized neuron, which is expected to respond optimally to applied inputs. A neural network is a collective set of such neural units, where individual neurons are connected through complex synaptic associations represented by weight coefficients. Each neuron contributes to the computational properties of the entire system. Convolutional Neural Networks (CNNs) can successfully capture spatial and temporal dependencies in images using appropriate filters. Due to the reduced number of parameters and the reusability of weights, CNNs are well-suited for image datasets. The network can be trained to better recognize the complexities of images. This paper highlights the strength of CNNs over traditional neural networks (NNs). CNNs achieved a multi-class classification accuracy of almost 97%, whereas traditional NNs achieved a binary classification accuracy of about 94% using the trainbr training function. Therefore, it is evident that CNNs and traditional NNs present similar results, with CNNs having a slight edge in performance.