Publications scientifiques
Enow Takang Achuo Albert
Ngonkeu Mangaptche Eddy Léonard
1. INTRODUCTIONThe continuous pursuit of better image classification AI models is crucial for several reasons. (1) Improved Accuracy: The primary goal of developing better models is to improve accuracy. Higher accuracy means the model can correctly classify more images, which is beneficial in many applications, including plant disease recognition. For instance, a more accurate model could lead to earlier detection of diseases [1]. (2) Efficiency: Better models can also be more efficient, requiring less computational resources or less time to train or make predictions. This is particularly important in real-time applications, such as image/video processing or in devices with limited computational power [2]. (3) Robustness: Improved models can be more robust to variations in the input data, such as changes in lighting, viewpoint, or scale. This is crucial for many real-world applications where these conditions can’t be controlled [3]. (4) Generalization: Better models can generalize more effectively to new, unseen data. This is important because a model’s ultimate purpose is to make accurate predictions on new data, not just the data it was trained on [4]. (5) Interpretability: As AI models are increasingly used in decision-making, it’s important that they’re interpretable, meaning that their predictions can be understood and explained by humans. Better models can provide more interpretable predictions, increasing trust and acceptance of AI systems [5]. (6) Fairness: Improved models can also be designed to be fairer, reducing bias in their predictions. This is particularly important in applications where biased predictions can have serious consequences [6]. Further to the foregoing, we set out to present an improved feature extraction ConvNet for better plant disease recognition (calledDetection ).
