MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications. Leaf image is captured and proposed to determine the health status of each plant. The acquired leaf images are converted into HSI format. On the data set of tomato and cucumber leaf diseases, the proposed model outperformed the traditional approaches in terms of the classification accuracy. Real time specific weed discrimination technique using multilevel wavelet decomposition was proposed by Siddiqil et.al. 142, 369–379. Each column of the confusion matrix stands for the number of instances in a ground truth class while each row stands for the number of instances in a predicted class to see if the system is confusing two classes. On one hand, the parameters of the first convolutional layer was reduced from 1,792 to 283, and the parameters of the second convolutional layer was reduced from 36,928 to 4,736, which contributes to reduce the consumption of computing resources and improve the generalization performance. Intell. (2014). The visualization of heatmaps of class activation refers to the production of heatmaps of class activation over input images (Selvaraju et al., 2017). Result of some plant leaves are shown below. Leaf blight produces dark brown patches on the surface of grape leaves. Adaptive moment estimation (Adam) was applied instead of Stochastic gradient descent (SGD), a traditional algorithm, as the optimization algorithm of the model. Densely Connected Convolutional Networks. 178, 7–11. Assume that h represents the height and P(x,y) is an arbitrary point in the image. Abadi, M., Barham, P., Chen, J., Chen, Z., Davis, A., Dean, J., et al. Limited by the number of images of the grape leaf disease data set, the model with a large size is prone to overfitting during the training process. doi: 10.3964/j.issn.1000-0593(2019)06-1864-06, Zhang, S., Wang, Z. Plant disease diagnosis is an art as well as science. doi: 10.1109/CVPR.2015.7298594, Szegedy, C., Vanhoucke, V., Ioffe, S., Shlens, J., Wojna, Z. Yinmao Song, ZhihuaDiao, Yunpeng Wang, Huan Wang developed feature extraction methods of crop disease based on computer image processing technology. Producing Grape is a daunting task as the plant is exposed to the attacks from various micro organisms ,bacterial diseases and pests .The symptoms of the attacks are usually distinguished through the leaves ,stems or fruit inspection . With those features, CNN-based models realize excellent recognition performance on grape leaf diseases. Although the accuracy of the model is slightly lower than before, the model can still accurately classify grape leaf diseases. However, current research involves using bacteriophages (viruses that kill bacteria) to stop and prevent the spread of Pierce’s Disease on wine grapes. Based on the Inception structure, the disease features in the original image can be extracted from multiple dimensions. This facilitates understanding of how successive convolution layers transform their input and of the meaning of each filter. Hence, various spectroscopy techniques have been widely applied in plant disease diagnosis and monitoring. In addition, accuracy curves were used to visually represent the accuracies and convergence speeds of the models. IDENTIFICATION OF NUTRIENT DEFICIENCIES FROM LEAF SYMPTOMS. Then image-processing techniques are applied to the acquired images to extract useful features that are necessary for further analysis. Plant Sci. The visualization of intermediate activation refers to the display of feature maps which are output by all kinds of convolution and pooling layers in the network for a specified input. In contrast, the CNN-based approaches extract the best classification features automatically. Comput. As the disease evolves, the leaves twist alongside the veins. Grapes Seasonal Calendar 5 DISEASES Leaf Unfolded Leaf Development Early Flowering Fruit Set/Swell Fruit Development Veraison Preharvest EASTERN GRAPES Powdery mildew Downy mildew Black rot Inspire Super and Quadris Top are not currently registered for sale or use on grapes in all states. proposed a technique for identification of grape disease through the leaf texture analysis and pattern recognition. In addition, a dense connectivity strategy is introduced to encourage feature reuse and strengthen feature propagation. doi: 10.3964/j.issn.1000-0593(2019)06-1898-07, Keywords: grape leaf diseases, convolutional neural networks, deep learning, image augmentation, disease identification, Citation: Liu B, Ding Z, Tian L, He D, Li S and Wang H (2020) Grape Leaf Disease Identification Using Improved Deep Convolutional Neural Networks. Some are interspecific crosses between American Vitis and Vitis vinifera. from our awesome website, All Published work is licensed under a Creative Commons Attribution 4.0 International License, Copyright © 2020 Research and Reviews, All Rights Reserved, All submissions of the EM system will be redirected to, International Journal of Innovative Research in Computer and Communication Engineering, Creative Commons Attribution 4.0 International License. Many grapes for the home garden are American grape varieties, largely because they are more disease resistant. Hence, it is challenging to comprehend the massive number of parameters, the multi-layer hidden structure, and other factors of these models. (2018). Precision, Recall and F1 Score are derived from the number of false positive (FP), true positive (TP), false negative (FN), and true negative (TN) results. The grape industry is one of the major fruit industries in China, and the total output of grapes reached 13.083 million tons in 2017. The last module is composed of two max-pooling layers, an Inception layer, a global average pooling (GAP) layer, and a 7-way Softmax layer. (2019). In this histogram equalization is used for pre-processing, features are extracted from wavelet decomposition and finally classified by Euclidean distance method [10]. J. Netw. In (Mohanty et al., 2016; Zhang and Wang, 2016; Lu J. et al., 2017; Lu Y. et al., 2017; Khan et al., 2018; Liu et al., 2018; Geetharamani and Pandian, 2019; Ji et al., 2019; Jiang et al., 2019; Liang et al., 2019; Oppenheim et al., 2019; Pu et al., 2019; Ramcharan et al., 2019; Wagh et al., 2019; Zhang et al., 2019a; Zhang et al., 2019b; ), CNNs are extensively studied and applied to the diagnosis of plant diseases. 2017M613216, by the Fundamental Research Funds for the Central Universities under Grant No. Libo Liu and Ouomin Zhou [4] studied the identification method of rice leaf disease according to the colour characteristics of leaf lesion area. 1–9. Agric. Electron. Variable Selection and Training Set Design for Particle Classification using a Linear and a Non-Linear Classifier. 201910712048. 76, 323–338. Energies 12, 1–11. This loss of features severely affects the model’s recognition accuracy. Additional configuration parameters are listed in Table 4. Due to the weak interpretive performance, the features that are learned by CNN-based models are difficult to represent in a human-readable form. Plant Disease Recognition Using Fractional-order Zernike Moments and SVM Classifier. The disease Powdery mildew originates from inoculum that overwinters in infected buds. Agric. First, the digital images are acquired from the environment using a digital camera. Rethinking the Inception Architecture for Computer Vision. The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. The second module, namely, the “cascade dense Inception module,” is composed of four Inception structures with dense connections. (A) The original image, (B) high brightness, (C) low brightness; (D) high contrast; (E) low contrast; (F) high sharpness; (G) low sharpness; (H) 90 degree rotation; (I) 180 degree rotation; (J) 270 degree rotation; (K) vertical symmetry; (L) horizontal symmetry; (M) Gaussian noise, and (N) PCA Jittering. Akbarzadeh, S., Paap, A., Ahderom, S., Apopei, B., Alameh, K. (2018). However, crop losses can be minimized, and specific treatments can be tailored to combat specific pathogens if plant diseases are correctly diagnosed and identified early. Copyright © 2020 Liu, Ding, Tian, He, Li and Wang. proposed a united convolutional neural networks architecture based on an integrated method. Neural Comput. In (Hamuda et al., 2017), Hamuda et al. Res. Confusion matrix, as a standard format for expressing accuracy evaluation, is expressed by matrix form with n rows and n columns. Process. Int. The performance of the model in the extraction of features at various scales has a substantial impact on the final recognition accuracy.