Performance Analysis of a VGG based Deep Learning Model for Classification of Weeds and Crops

VGG19 is a deep convolutional neural network designed for image classification tasks and has attained cutting-edge performance on a number of benchmark datasets. However, adding additional fully connected dense layers can improve the model with additional capacity to learn more complex representations of the input data. The proposed modification involves adding two completely coupled dense layers, after the last convolutional layer, each with a SoftMax activation function. The first dense layer has 512 neurons, while the second dense layer has 256 neurons. After the two dense layers, a layer of dropouts is added with a rate of 0.5 to minimize overfitting. The proposed modification to the VGG19 architecture has shown improved performance on crop and weed classification tasks, achieving higher accuracy than the original VGG19 model. The added layers provide additional capacity to the model, allowing it to learn more complex representations of the input data, while the dropout layer helps to prevent overfitting and improve generalization. Accuracy of modified VGG architecture is about 96% while that of the existing CNN architecture is about 88%. Hence the proposed model using the modified VGG architecture obtained 8% higher accuracy than the existing CNN model and standard VGG model.