Weed management is one of the most important aspects of crop productivity; knowing the amount and the locations of weeds has been a problem that experts have faced for several decades. This paper presents three methods for weed estimation based on deep learning image processing in lettuce crops, and we compared them to visual estimations by experts. One method is based on support vector machines (SVM) using histograms of oriented gradients (HOG) as feature descriptor. The second method was based in YOLOV3 (you only look once V3), taking advantage of its robust architecture for object detection, and the third one was based on Mask R-CNN (region based convolutional neural network) in order to get an instance segmentation for each individual. These methods were complemented with a NDVI index (normalized difference vegetation index) as a background subtractor for removing non photosynthetic objects. According to chosen metrics, the machine and deep learning methods had F1-scores of 88%, 94%, and 94% respectively, regarding to crop detection. Subsequently, detected crops were turned into a binary mask and mixed with the NDVI background subtractor in order to detect weed in an indirect way. Once the weed image was obtained, the coverage percentage of weed was calculated by classical image processing methods. Finally, these performances were compared with the estimations of a set from weed experts through a Bland–Altman plot, intraclass correlation coefficients (ICCs) and Dunn’s test to obtain statistical measurements between every estimation (machine-human); we found that these methods improve accuracy on weed coverage estimation and minimize subjectivity in human-estimated data.

Each crop has their own weed problems. Therefore, to understand each problem, agronomists and weed scientists must be able to determine the weed abundance with the most precise method. There are several techniques to scouting, including visual counting for density or estimations for coverage of weeds. However, this technique depends by the evaluator subjectivity, performance, and training, causing errors and bias when estimating weeds abundance. This chapter introduces a methodology to process multispectral images, based on histograms of oriented gradients and support vector machines to detect weeds in lettuce crops. The method was validated by experts on weed science, and the statistical differences were calculated. There were no significant differences between expert analysis and the proposed method. Therefore, this method offers a way to analyze large areas of crops in less time and with greater precision.

In Colombia, peas are the second most important legume after the bean, and weeds are the main biotic factor that limits production, causing losses of up to 100%. Manual control can represent up to 40% of the labor force. The critical period in the crop-weed competition is the first third of the crop cycle; therefore, pre-emergent herbicide applications are a cost-effective way to control weeds. Common variables for assessing weed-control efficacy include, weed density (individuals/area), which is precise but time consuming, and weed coverage (%), which is faster but very subjective. Therefore, pre-emergence herbicides and a weed-control evaluation method that standardizes, facilitates, and provides greater precision are needed for peas cultivation and experimentation. Five pre-emergent herbicides (linuron, S-metolachlor, metribuzine, oxifluorfen and pendimetalin) were evaluated at two doses in a greenhouse pea crop. Also, two methods (quantification process of multispectral images and conventional human visual) for assessing weed coverage and control efficacy were compared. The best herbicide treatment for the dry grain yield was metribuzine (2.36 t ha-1). Furthermore, the effectiveness of the weed control was 88% at 36 days after sowing, which is optimal. Finally, there was agreement between the weed assessment methods (human vs. machine). The intraclass correlation coefficient was over 0.95, which validates the use of machine quantification for weed coverage.

El presente capítulo tuvo como propósito analizar la disrupción de la exclusión social a la discapacidad a través de las tecnologías asistivas. El estudio se ubicó en el paradigma cualitativo a través de un estudio documental, con diseño bibliográfico descriptivo. Para lograr lo anterior expuesto, las categorías analizadas fueron: Aproximaciones de la exclusión social, Tecnologías asistivas como modelo de apoyo a los procesos inclusivos, Tecnología asistiva de asistencia integral y Tecnología asistiva de uso específico para construir un análisis sobre la base de las teorías planteadas por Casado (1976), Foucault (1974), Lenoir (1974), Tezanos (1999), Borg, Lindström y Larsson (2009), Molero et al. (2020), y WHO, World Health Organitation, (2020), entre otros. Dentro de los hallazgos se presentaron una serie de estudios aplicados en tecnologías asistivas que permitieron evidenciar la disrupción de la exclusión social a la discapacidad a través de estas herramientas. Como conclusión, los adelantos en materia de tecnología de asistencia que ha traído consigo avances sustanciales en armonía a los nuevos paradigmas así como la participación activa de organismos internaciones como la Organización Mundial de la Salud, quien delinea la hoja de ruta en relación a la discapacidad, sientan las bases para que este tipo de herramientas emerjan como instrumentos disruptivos a la exclusión social en diferentes dinámicas aumentando su pertinencia en los contextos globales.

This chapter presents a study to identify with classification techniques and digital recognition through the construction of a prototype phase that predicts criminal behavior detected in video cameras obtained from a free platform called MOTChallenge. The qualitative and descriptive approach, which starts from individual attitudes, expresses a person in his expression, anxiety, fear, anger, sadness, and neutrality through data collection and feeding of some algorithms for assisted learning. This prototype begins with a degree higher than 40% on a scale of 1-100 of a person suspected, subjected to a two- and three-iterations training parameterized into four categories—hood, helmet, hat, anxiety, and neutrality—where through orange and green boxes it is signaled at the time of the detection and classification of a possible suspect, with a stability of the 87.33% and reliability of the 96.25% in storing information for traceability and future use.