Early Disease Classification of Mango Leaves Using Feed-Forward Neural Network and Hybrid Metaheuristic Feature Selection

“After performing contrast enhancement step, we investigate four popular CNN models, namely AlexNet [30], VGG16 [31], ResNet [32] (ResNet-50 variant).”

“After performing contrast enhancement step, we investigate four popular CNN models, namely AlexNet [30], VGG16 [31], ResNet [32] (ResNet-50 variant).”

“APPROACH WITH CNNS After performing contrast enhancement step, we investigate four popular CNN models, namely AlexNet [30], VGG16 [31], ResNet [32] (ResNet-50 variant).”

“Moreover, the optimization algorithm for the training process is done by Adam optimizer[36].”

“In this study, we implement a wrapper-based method with a hybrid meta-heuristic called Adaptive Particle – Grey Wolf Optimization (APGWO)[39], which is combined from Particle Swarm Optimization (PSO)[40] and Grey Wolf Optimization (GWO) [22].”

“Secondly, when testing with images taken under conditions that are different from the training images, the CNN models’ performance decreased signifi-cantly, as stated by Mohanty et al. [7].”

“They consist of multiple layers with later ones built on top of previously learned features [3].”

“Contrast Limited Adaptive Histogram Equalization (CLAHE) was proposed by K. Zuiderveld in 1994 [37].”

“Konstantinos et al. [5] implemented a Visual Geometry Group (VGG) model for plant disease detection, in which the network achieved 99.53% of accuracy over the data given in [1].”

“Some examples are Recursive feature elimination [17], Sequential feature selection algorithms [18], and Genetic algorithms [19].”

“[8] conducted a review of the performances of different deep learning models namely Visual Geometry Group (VGG), Inception V4, ResNet, and DenseNet in disease classification using the ‘‘Plant Village’’ dataset, in which DenseNet was the most efficient with 99.”

“Most of them make use of the so-called ‘‘Plant Village’’ [1] dataset, a widely known dataset that is available online, with CNNs being by far the most popular models.”

“…feature selection algorithms were also introduced in many previous works to select a subset of features, including binary grey wolf optimization (BGWO) [23], binary particle swarm optimization (BPSO) [24], ant colony optimization (ACO) [25], and binary differential evolution (BDE) [26].”

“…of disease in plants such as self-organizing feature map, backpropagation algo-rithm, support vector machines (SVMs), etc. Singh et al. [12] used ANN together with image segmentation to detect diseases on various types of plants, namely banana, beans, jackfruit, lemon, mango, potato,…”

“[4] conducted a review of plant disease detection and classification by Deep Learning techniques.”

“Khirade et al. [11] conducted a review of various techniques to segment the infested part of the plant.”

“However, the CNNs require a large amount of data for their training [2].”

“[6] used AlexNet and VGG16 to classify tomato leaf diseases, in which VGG16 reached 97.”

“…feature selection algorithms were also introduced in many previous works to select a subset of features, including binary grey wolf optimization (BGWO) [23], binary particle swarm optimization (BPSO) [24], ant colony optimization (ACO) [25], and binary differential evolution (BDE) [26].”

“Some examples are Recursive feature elimination [17], Sequential feature selection algorithms [18], and Genetic algorithms [19].”

“A new dataset [14] was used in [10]”

“…feature selection algorithms were also introduced in many previous works to select a subset of features, including binary grey wolf optimization (BGWO) [23], binary particle swarm optimization (BPSO) [24], ant colony optimization (ACO) [25], and binary differential evolution (BDE) [26].”

“Many evolutionarymetaheuristics-based feature selection methods are also proposed, many of them are wrapper-type since it has been proven that wrapper provides better performance [20].”

“Transfer learning (TL) is a solution for the lack of training data in deep learning [33].”

“this method has shown effectiveness in image classification task [7], [34], [35].”

“[9] to classify tomato diseases, which achieved 88% accuracy.”

“[21] proposed a competitive binary Grey Wolf Optimizer (CBGWO), which is based on the Grey Wolf Optimizer proposed by Mirjalili et al.”

“this method has shown effectiveness in image classification task [7], [34], [35].”

“Some examples are Recursive feature elimination [17], Sequential feature selection algorithms [18], and Genetic algorithms [19].”

“…feature selection algorithms were also introduced in many previous works to select a subset of features, including binary grey wolf optimization (BGWO) [23], binary particle swarm optimization (BPSO) [24], ant colony optimization (ACO) [25], and binary differential evolution (BDE) [26].”

“Many contrast enhancement methods have been widely applied to improve the quality of the image [28].”

“A new dataset [14] was used in [10] to detect mango leaves’ diseases, however, they only focus on healthy leaves and those infected with ‘‘Anthracnose’’ only.”

“Details of the functions can be found Algorithm 1 Proposed APGWO Initialize the particle population Initialize parameters while (t Max number of iteration) for each particle with position x p calculate fitness value f(x p ) if f(x p ) is better than pbest p then pbest p ← x p f(pbest p ) is better than gbest then gbest…”

“Pham et al. (2020) used an ANN model to classify mango leaves in the PlantVillage dataset, achieving an 89.41% accuracy rate.”