An efficient densely connected convolutional neural network for identification of plant diseases

In this research work, novel densely connected convolutional neural network (DCCNN) based deep learning architectures are proposed to identify diseases in the apple, corn, cucumber, grape and potato plant leaves. The three concrete novel deep learning architectures, namely 6 block DCCNN, 7 block DCCNN and 8 block DCCNN are compared with state-of-the-art conventional machine learning and deep learning approaches. The performance is evaluated using training accuracy, validation accuracy, loss values, confusion matrices, sensitivity, specificity, precision and F-score measures. The 8 block DCCNN achieved greater identification accuracy of 99.78%, 98.85%, 98.23%, 99.78, % and 99.83% for apple, corn, cucumber, grape and potato plant leaf dataset respectively. The higher identification accuracy is achieved in the proposed 8 block DCCNN since the densely connected convolution neural network layers are incorporated with modified dense blocks.