Identification of different plants, weeds, or any related type of vegetation is an important aspect of agricultural robotics and technologies. With the help of image processing and computer vision, multiple attempts have been made to achieve these results. These approaches made use of the shape and color of the leaf to identify a particular plant. But it can be observed that this approach has some limitations resulting in false positive and true negative errors. To overcome these limitations, the authors propose a novel approach of using Laplacian filter to extract veins morphology of leaves of a plant. This veins pattern is unique to every plant. With this Laplacian filter and data augmentation techniques, a unique dataset is developed on which a deep learning model can be trained. Based on this approach, the proposed system applies a deep learning algorithm called YOLO for plant identification. After preprocessing and YOLO training, the model is able to distinguish between plant and a weed successfully and create a bounding box for the detected type of plant.
Top2. Literature Survey
Recently, there has been a good amount of research and development in this sector, some of which is explained briefly below.
The aim of this study is to apply image processing techniques for agricultural plant identification, which can lead to faster and more efficient specimen identification and classification (Sonnad, et al., 2022). The proposed method uses various factors such as colour moments, vein features, texture features based on lacunarity, GIST, Local Binary Pattern (LBP), and geometric features, among others, to identify plant leaf photos. The features are normalised, and the ‘Pbest-guide binary particle swarm optimisation (PBPSO)’ technique is used to reduce the features. Several machine learning classifiers are evaluated, including multi-SVM, k-nearest neighbour, decision trees, and naive Bayes, and the decision tree classifier showed the best performance. The proposed method achieved an accuracy of 98.58% and 90.02% for the “Flavia” and “Folio” datasets, respectively (Keivani, et al., 2020).
The detection of invasive species in ornamental lawns and sports turf can be accomplished through the use of edge detection algorithms. In this study, 12 different edge detection filters were evaluated on collected photos. To minimise false positives, the outputs of the three most effective filters - sharpening (I), sharpening (II), and Laplacian - were combined using various cell values. Two filters were then selected for further investigation based on tests conducted with different cell sizes. To identify the optimal cell size and the most effective filter, box plots were used. The best results were obtained by applying the sharpening (I) filter with a cell size of 10 and the aggregation method with the minimum value. Finally, based on the number of false positives, false negatives, and generated indexes, a threshold value was selected to achieve the best performance in terms of Precision, Recall, and F1-Score. The results varied slightly for sports grass and ornamental turf (Parra, et al., 2020).