Intelligent Plant Leaf Disease Detection System using Image Processing Techniques

Abstract

Plant leaf disease detection plays a crucial role in agricultural systems as it enables early identification and effective management of diseases, leading to improved crop yield and reduced economic losses. newlineThis research work provides a solution for lower feature discrimination, problem with data imbalance, larger intra-class variation, lower inter-class variation, low recognition rate, and a poor balance between recall and accuracy. proposed work will be carried out in two stages, the first is creation of a Generative Adversarial Network for the purpose of data augmentation and a DCNN for the purpose of plant leaf disease detection. In the first step, we will detect tomato plant leaf disease by implementing a lightweight parallel deep convolutional neural network, also known as LPDCNN. This will boost the feature distinctiveness while minimizing the difficulty of filter size selection. newlineDCNNs and Generative Adversarial Neural Networks (GAN) are going to be used in the initial studies in order to address the detection of plant leaf illnesses utilizing these two types of neural networks as a representation of a generalized solution for the identification of flaws in a variety of diseases. The DCNN provides more feature distinctiveness and correlation, while the GAN-based architecture helps to reduce the problem of class imbalance that arises as a result of an uneven quantity of samples being taken from various classes during training. This problem arises because unequal samples taken. An examination is carried out by making plant village dataset, to see whether the projected system will be capable of successfully identifying a number of defects. It has been observed that the proposed combination of GAN and DCNN provides an accuracy of 99.74%, a recall rate of 0.99, a precision of 0.99, and an F1- score of 0.99. newlineThe next research work explores a lightweight parallel Deep Convolutional Neural Network (LPDCNN) with the intention of determining the cause of tomato plant leaf disease. This DCNN increases the distinctness of the features while also reducing the complexity of selecting the appropriate filter size. In addition, a cyclic generative adversarial network for the production of synthetic pictures has been successfully developed. This implementation helps to alleviate the class imbalance problem that was produced by uneven samples in the training dataset. The LPDCNN that has been proposed helps to improve the feature representation of leaf images and helps to raise the PLDD accuracy for a range of diseases. These benefits may be realized by using the LPDCNN. newlineviii newlineWhen using the recommended LPDCNN, the accuracy of disease detection for tomato PLDD at two classes, six classes, and ten classes, respectively, is 99.14%, 99.05%, and 98.11%, respectively, when utilizing the dataset from Plant Village. When it comes to illness identification, the LPDCNN-CGAN that was suggested exhibits a 2.15% gain in accuracy over the LPDCNN even though no more data was augmented. This improvement was measured using 10-class disease detection. When compared to the traditional state of the art, the recommended method has less trainable parameters (188426) for the 9 class PLDD. This represents a significant advancement over the prior state of the art and contributes to an increase in the possible implementation flexibility on independent devices. newline