An experimental investigation to spotting the weeds in rice field using deepnet

Abstract

This research article mainly focuses on the process of detecting unwanted plants, which grows in the agricultural field. The backbone of our nation is agricultural, but due to the usage of pesticides to remove the weeds, could cause the reduction in yield, apart from the climatic changes. Henceforth it is needs an optimized system on the detection of weed without affecting the crop. The proposed model gives the suitable solutions to this issue. Fundamentally this idea is based on the accessing the field with the machine learning technique and deep neural network. The ultimate functionality of the neural network defines the level of prediction of the image as weed or crop. This process is tedious when the crop and weeds are in marginal level on the field. Our model analyzes these cases and predicts accuracy of 94% in fast region convolution neural network, 92% and 86% in Region convolution neural network and convolution neural network. This model is contrasted with other models such as support vector machine, convolution neural network, artificial neural network and the evaluation parameters considered are specificity, precision, recall and F1 score. The validation and training sets given to the model and the accuracy levels are tabulated for each type of weeds, especially connected to rice field. In addition, the hardware primary factors of processors such as power, area, latency and error occurrence has compared in different adder structures after implementing these models on it. The adder models used for analyse is carry cutback adder, carry latency adder under the approximate and accuracy model and the results shows the optimized adder structure for this applications based on low power, error occurrence and low latency.