Flora Maria de Melo Villar

Coffee production in Brazil requires efficient monitoring tools, especially in regions with heterogeneous terrain and planting patterns. Accurate delineation of coffee canopies from UAV imagery is essential for extracting reliable agronomic information and supporting precision agriculture practices. Multispectral UAV images were acquired over a commercial coffee farm in the Matas de Minas region, Brazil. An orthomosaic was generated and manually annotated to create binary ground-truth masks of coffee canopies. A U-Net convolutional neural network was trained using multispectral bands and the Normalized Difference Vegetation Index (NDVI) as input. Model performance was evaluated using Dice coefficient, Intersection over Union (IoU), precision, recall, and accuracy. The proposed model achieved a Dice coefficient of 0.89, IoU of 0.83, and overall accuracy above 0.92. The predicted coffee canopy area (1.50 ha) was consistent with the reference area obtained from manual annotation (1.45 ha), indicating reliable canopy delineation under heterogeneous field conditions. The results demonstrate that semantic segmentation using multispectral UAV imagery and a U-Net model is an effective and applicable approach for coffee canopy mapping, with potential use in precision agriculture and farm-scale decision-support systems.

Coffee production in mountainous regions faces significant challenges to mechanization, particularly in management and fertilization. Fertilizer application remains largely manual, reducing accuracy and failing to meet the demands of variable-rate application (VRA). This study developed a portable VRA fertilizer applicator with an embedded electronic control system. The innovation lies in its electrically driven metering mechanism integrated with an electronic control unit (ECU), enabling site-specific fertilization based on prescription maps or predefined rates while recording application coordinates. The mechanism was tested under laboratory and field conditions, evaluating its performance across four fertilizer types, varying inclination angles, and rotational speeds. Results showed a coefficient variation of 0.41% for doses above 24 g, demonstrating high consistency irrespective of fertilizer type or terrain slope. Field tests using potassium chloride (KCl) prescriptions (55, 123, and 185 g/plant; 220, 492, and 740 kg/ha) revealed minimal deviations, with the largest at 22.72 g and the smallest at 0.384 g. These findings demonstrate the applicator’s precision and efficiency, addressing the challenges of mountainous terrains. This system provides technological advancement for sustainable coffee production, enhancing resource optimization and supporting precision agriculture in challenging environments.

The development of autonomous agricultural robots using a global navigation satellite system aided by real-time kinematics and an inertial measurement unit for position and orientation determination must address the accuracy, reliability, and cost of these components. This study aims to develop and evaluate a robotic platform with autonomous navigation using low-cost components. A navigation algorithm was developed based on the kinematics of a differential vehicle, combined with a proportional and integral steering controller that followed a point-to-point route until the desired route was completed. Two route mapping methods were tested. The performance of the platform control algorithm was evaluated by following a predefined route and calculating metrics such as the maximum cross-track error, mean absolute error, standard deviation of the error, and root mean squared error. The strategy of planning routes with closer waypoints reduces cross-track errors. The results showed that when adopting waypoints every 3 m, better performance was obtained compared to waypoints only at the vertices, with maximum cross-track error being 44.4% lower, MAE 64.1% lower, SD 39.4% lower, and RMSE 52.5% lower. This study demonstrates the feasibility of developing autonomous agricultural robots with low-cost components and highlights the importance of careful route planning to optimize navigation accuracy.

It is common to observe conventional methods for estimating soybean crop yields, making the process slow and susceptible to human error. Therefore, the objective was to develop a model based on deep learning to estimate soybean yield using digital images obtained through a smartphone. To do this, the ability of the proposed model to correctly classify pods that have different numbers of grains, count the number of pods and grains, and then estimate the soybean crop yield was analyzed. As part of the study, two types of image acquisition were performed for the same plant. Image acquisition 1 (IA1) included capturing the images of the entire plant, pods, leaves, and branches. Image acquisition 2 (IA2) included capturing the images of the pods removed from the plant and deposited in a white container. In both acquisition methods, two soybean cultivars, TMG 7063 Ipro and TMG 7363 RR, were used. In total, combining samples from both cultivars, 495 images were captured, with each image corresponding to a sample (plant) obtained through methods AI1 and AI2. With these images, the total number of pods in the entire dataset was 46,385 pods. For the training and validation of the model, the data was divided into subsets of training, validation, and testing, representing, respectively, 80, 10, and 10% of the total dataset. In general, when using the data from IA2, the model presented errors of 7.50 and 5.32% for pods and grains, respectively. These values are considerably lower than when the model used the IA1 data, where it presented errors of 34.69 and 35.25% for pod and grain counts, respectively. Therefore, the data used from IA2 provide better results to the model.

Crop productivity estimate aims at the economic definitions about crop, agricultural management, and land use, among others. However, it is common to observe the use of visual methods to estimate the productivity of the soybean crop through the classification of pods, resulting in a slow, costly method besides being susceptible to human errors. Thus, the objective of this work was to carry out the training of two deep learning methods to classify soybean pods according to the number of grains based on images obtained using a smartphone. Data collection was carried out at the Federal University of Viçosa (UFV). Data consisted of capturing images from a smartphone and training two deep learning models: Mask R-CNN and YOLOv4. To capture the images, the soybean pods were pulled from the plants and placed in a white-bottom container. This procedure occurred for each plant collected. Both models tended towards a better classification for the two- and three-grain pods, reaching a value of 90% for the F1 score metric. This fact may have occurred because of the greater amount of these two types of pods present in the chosen cultivars. Finally, the potential of using deep learning to classify soybean pods based on the number of grains was observed.

The macaw palm (Acrocomia aculeata) is a palm tree native to tropical forests that stand out due to its great potential for oil production. This study was developed with the objective of constructing a high-fidelity model of the macaw palm fruit-rachilla system for the purpose of simulating its dynamic behavior when subjected to mechanical vibrations. The finite element method was used to determine the natural frequencies and modes of vibration of the system. The three-dimensional models of the fruit-rachilla systems were elaborated using CAD3D Fusion 360 software. The modal properties of the fruit-rachilla systems were obtained based on the models developed by varying the elasticity modulus values of the system. The parameters of greatest influence in the estimation of natural frequencies are the elasticity modulus, especially that of the fruit-rachilla joint, and the specific mass. The models that take into account the three-dimensional strains along the rachilla are the least sensitive to variations in the mechanical properties (elasticity modulus and specific mass) and are shown to be more representative of the actual physical system.

The macaw palm has been domesticated due to its potential use in the production of biofuel, in addition to several co-products that can be generated from its oil and pulp. One of the current challenges in this area is the harvesting, as there are no specific machines for this operation. Therefore, it is necessary to determine the appropriate information regarding the physical properties of the plant, so that it is feasible to develop the technologies necessary for the commercial scale application of macaw palm, allowing it to contribute to the sustainable production of raw material for the biofuel industry and other co-products. The principle of mechanical vibration can be used to shed fruit from trees when ripe, and it can be a method used for harvesting. Thus, as proposed in this study, it was necessary to study the dynamic behavior of the fruit-rachilla system during vibration. Hence, the modal properties of the system were determined. A study on the dynamic behaviors was carried out using a deterministic finite element model, and the natural frequencies were obtained through a frequency-scanning test to evaluate the model. The mean relative error (MRE) between the measured and simulated natural frequencies was also used to evaluate the model. The natural frequencies, determined experimentally, varied from 26.21 to 33.45 Hz on average, whereas the simulated frequencies varied from 24.81 to 39.27 Hz. The overall MRE was 9.08%. Once the model was validated, a sensibility test was carried out, which showed that the density of fruit and the elasticity modulus are the parameters that most influence the natural frequencies of the fruit-rachilla system.

This study evaluated the use of low-cost materials for radiometric calibration of multispectral images. Four materials were tested: plywood panels painted with matte paint (M1); plywood panels covered with synthetic nappa leather (M2); Ethylene Vinyl Acetate (EVA) panels (M3), and plywood panels covered with Polyvinyl chloride (PVC) canvas (M4). The useful life of all materials and the errors associated with the calibration was determined. The M1 and M2 panels presented the lowest errors (RMSE), whereas the M4 panels showed the highest errors. Finally, the M3 panels showed the lowest resistance to use, whereas the M1 panels showed the greatest durability.