An exhaustive research was done on several papers describing various methods adopted for weed detection. These papers were summarized as follows.
Image processing techniques and machine vision are broadly used in various fields such as agriculture industry or for detection of an object. The images are mathematically represented as rows and columns with red. So, automatic weed management systems are impacting the economy of the country green and blue channels. The weed-infected crop images are inspected either manually or automatically with robotics. Sometimes the weed and crops are of same color, in this condition we use the fuzzy image processing for oscillation basis. The absence of weeds in crops is detected using the advanced fuzzy set algorithms. The starting endeavors to distinguish weed seedlings using machine vision were centered on geometrical estimations such as shape, angle proportion, length region etc. Afterwards, color-based pictures were successfully used to identify weed infected area. The weed scope and weed patchiness are based on the computerized images where employing fuzzy-based calculations is used to select the site of the weed area. Tellaeche et al. (2007) proposed to apply k means and Bayesian-based algorithms for decision-making regarding spraying of adequate amount of herbicides. Fast Fourier transform process was applied for weed detection in various corn fields (Nejati et al., 2008). Agriculture industry plays one of the most essential roles in economy of any country. These days, weeds are controlled by automatic robotic cultivators. The robotics captures the images of the crops and defines the weed-infected region. Then it eliminates the weed by spraying herbicide accurately on the weed. Fast Fourier transform algorithm is used in weeding robot to detect the accurate area of the weed. The process includes preprocessing, frequency and density filtering, and finally post-processing. In preprocessing, separation of background is done through Euclidean distance based on green and blue pixel channels. Density and frequency features are collected from preprocessed images and clustered by optimization classifiers. Finally, the weed area foreground and background are distinguished accurately.
Bhongale and Gore (2017) proposed a dilation and erosion method applied to segmentation process, it is used to classify the weeds in crops. In this method, a threshold value is applied to segment the weed area. These kinds of algorithms are used in agricultural and crop scouting. In crop scouting, applications are used for pest and weed detection in various plants which reduces the manual working time. The image processing algorithms are used to improve the accuracy of weed area detection in broad and narrow leaves. Yield is lost due to weeds in the crop, so their detection by automatic process enhances the yield. Dyrmann et al. (2017) proposed to detect the weeds using fully convolution neural network. GoogleNet is one of the networks which can process thousands of images to detect even the overlapping weeds. The network produces the predictive convergence map and bounding boxes. The convergence maps show if the weeds are present in crops. Output convergence maps are used to locate the weeds in the image. The proposed work helped to overcome the problem of detection of small weeds.
Pulido Rojas et al. (2019) proposed an application named Auto weed used in Australian rangelands. This real-time application detects the weed area by collecting the dataset of hyper-spectral images and thus classifies the weed area. Such robust methods help in improving the accuracy of weed area detection. Lot of research work is going on in the robotic weed control field. Robotic weed control contains four processes: mapping, control, guidance and detection of weeds. Weed control robotics are developed based on spectrum, image and spectral image-based methods to detect the weeds using aerial and ground photography. The image features are taken in numerical format to check whether it is a crop or weed, this functionality is known as image analytics. The most important part is to detect the features and classify them according to various features like shape, texture, statistical features, etc. Weeds reduce the crop production, so various applications are implemented which work better and reduce the human working time. The weed locations are identified through robotics, so a continuous research is going on in this area. The machine learning algorithms are applied to detect the weed infected region and measure the quantity of herbicides required. These automatic methods can surely help to enhance the economic level of a country by increasing the agricultural production. If any weeds are not detected by four-wheel robots, they move forward and check other fields. Color-based image processing is suggested by most of the authors for weed detection because it can capture variations of pixels in red, green and blue ranges. Sometimes oscillation occurs in weed area classification, but this problem is solved by advanced fuzzy set theories. From this literature survey, various weed processing algorithms are evaluated. Linear searching algorithm is used to detect the weed rows in crop farming fields.
Leave a Reply