Cocoa Pods Diseases Detection by MobileNet Confluence and Classification Algorithms

• Published in: International journal of advanced computer science & applications Vol. 14; no. 9
• Main Authors: MAMADOU, Diarra, AYIKPA, Kacoutchy Jean, BALLO, Abou Bakary, KOUASSI, Brou Médard
• Format: Journal Article
• Language: English
• Published: West Yorkshire Science and Information (SAI) Organization Limited 2023; The Science and Information Organization
• Subjects: Algorithms; Artificial neural networks; Classification; Cocoa; Computer Science; Economic impact; Neural networks; Support vector machines
• ISSN: 2158-107X; 2156-5570; 2156-5570
• DOI: 10.14569/IJACSA.2023.0140937

Cocoa cultivation is of immense importance to the people of Côte d'Ivoire. However, this culture is experiencing significant challenges due to diseases spread by various agents such as bacteria, viruses, and fungi, which cause considerable economic losses. Currently, the methods available to detect these cocoa diseases force farmers to seek the expertise of agronomists for visual inspections and diagnostics, a laborious and complex process. In the search for solutions, many studies have opted for using convolutional neural networks (CNNs) to identify diseases in cocoa pods. However, an essential advance is to develop hybrid approaches that combine the advantages of a CNN with sophisticated classification algorithms. This research stands out for its innovative contribution, combining MobileNetV2, a convolutional neural network architecture, with algorithms, such as Logistic Regression (LR), K Nearest Neighbors (KNN), Support Vector Machines (SVM), XGBoost, and Random Forest. The study was conducted in two distinct phases. First, each algorithm was evaluated individually, and then performance was measured when MobileNetV2 was merged with the algorithms mentioned. These hybrid approaches complement and amplify MobileNetV2's capabilities. To do so, they draw on MobileNetV2's inherent capabilities to extract key features and enhance information quality. By combining this expertise with the classification methods of these other models, hybrid approaches outperform individual techniques. Accuracy rates range from 72.4% to 86.04%.This performance amplitude underlines the effectiveness of the synergy between the extraction characteristics of MobileNetV2 and the classification skills of other algorithms.