A Systematic Review of Different Categories of Plant Disease Detection Using Deep Learning-Based Approaches

Artificial intelligence has a significant impact on all sectors. It is revolutionizing agriculture by replacing traditional methods with more efficient techniques and helping the world improve. The primary purpose of the article is to develop a deep learning-based system to detect and classify plant diseases to save the crop and the time of farmers as well as to protect people from getting ill. Initially, the data was taken from the PlantVillage dataset in fifteen classes, such as tomato spider mates, pepper bell healthy, tomato late blight, pepper bell bacterial spot, tomato early blight, potato late blight, tomato bacterial spot, potato early blight, potato healthy, tomato septoria leaf spot, tomato leaf mold, tomato yellow leaf curl virus, tomato target spot, tomato mosaic virus, and tomato healthy. Later the data is pre-processed to remove the noisy signals, and the pixel values are added and removed to resize the image size using the dilation and erosion technique. For feature extraction, contour feature techniques have been used along with adaptive thresholding techniques to obtain the cropped image. For classification, ten deep learning models such as DenseNet201, DenseNet121, NasNetLarge, Xception, ResNet152V2, EfficientNetB5, EfficientNetB7, VGG19, and MobileNetV2 along with the hybrid model (EfficientNetB7 and ResNet152V2) have been applied. The models have been trained and later evaluated based on precision, recall, accuracy, F1-score, and loss. The performance of the models has also been assessed for different classes during the training and validation phases. While experimenting, it was found that DenseNet201 obtained the highest validation accuracy and loss by 98.67% and 0.04. For precision, recall, and F1 score, DenseNet201 again got the highest values of 0.98, 0.99, and 0.98, respectively.