Development of Modified CNN Algorithm for Agriculture Product: A Research Review

Abstract

Now a day, with the increase in world population, the demand for agricultural products is also increased. Modern days electronic technologies combined with machine vision techniques have become a good resource for precise weed and crop detection in the field. It is becoming prominent in precision agriculture and also supporting site-specific weed management. By reviewing as there are so many different kinds of weed detection algorithms that were already used in the weed removal process or in agriculture. By the comparative study of research papers on weed detection. In this paper, we have suggested advanced and improved algorithms which take care of most of the limitations of previous work. The main goal of this review is to study the different types of algorithms used to detect weeds present in crops for automated systems in agriculture. This paper used a method that is based on a convolutional neural network model, VGG16, to identify images of weeds. As the basic network, VGG16 has very good classification performance, and it is relatively easy to modify. Download the weed dataset. This image dataset has 15336 segments, being 3249 of soil, 7376 soybeans, 3520 grass, and 1191 broadleaf weeds. Our model fixes the first 16 layers of  VGG16 parameters for layer-by-layer automatic extraction of features, adding an average pooling layer, convolution layer, Dropout layer, fully connected layer, and softmax for classifiers. The results show that the final model performs well in the classification effect of 4 classes. The accuracy is 97.76 %. We will compare our result with the CNN model. It provides an accurate and reliable judgment basis for quantitative chemical pesticide spraying. The results of this study can provide an overview of the use of CNN-based techniques for weed detection.