SUMMARY
This paper exposes a comparative analysis of three weed classification strategies based on area and texture features over images of vegetable crops, focus on provide a technological tool to support farmers in their maintenance tasks. The classification alternatives embrace a basic approach which defines an umbral according to scene features, indeed, a detection with a certain degree of uncertainty on the decision region is purposed and a rigid boundary decision arrangement are exposed. A first mode carry out an unsupervised learning, it uses area and color features with a practical thresholding classifier to differentiate between weed and vegetable classes, the following two, extracts statistical measures of autocorrelation, contrast, correlation and others, from grey level co-occurrence matrices to calculate texture features, next, a principal component analysis is made for dimensionality reduction. These patterns serve as basis for training K-Nearest Neighbor and Support Vector Machine classifiers. The algorithms performance is measured calculating sensitivity (SN), specificity (SP), positive and negative predicted values (PPV and NPV), also, the execution time is stored and tabulated in order to evaluate the proposed methods. Finally, the results show a similar performance of correct classification over 90 and 80% on SN and SP indices respectively, however, approaches present a clear difference in execution time respect of train an evaluation stages.