Spilocaea oleagina is a common and dangerous fungal disease in olive trees that significantly reduces olive production. The early and accurate detection of this disease is essential for implementing effective control measures. In this study, we propose the creation of a new meta-architecture for identifying peacock spots on olive leaves. This new meta-architecture integrates Xception and VGG16 as the basis for the methodology employed in this study. Furthermore, a machine learning approach was used to pretrain the model in an unsupervised manner, thereby improving its generalization capacity. Metrics such as Kappa (K), True Skill Statistic (TSS), Proportion of Expected Success (Pe), Threat Index (Ts), and the Heidke Skill Score (HSS) were used to validate the model results. Based on these indicators, we evaluated the robustness, accuracy, and ability of the model to identify peacock spots. The testing results showed that the suggested meta-architecture, named SSL-XceVNet, substantially outperformed the baseline XceVNet model (88.24%) in detecting peacock spots, with an accuracy of 95.22%. The validation measurement results (K, 73.57%; TSS, 72.66%; Pe, 55.06%; Ts, 90.71%; HSS, 90.44%) attested to the efficacy and resilience of the suggested model, underscoring its ability to generalize and produce accurate predictions under a variety of circumstances. In conclusion, the combination of Xception and VGG16 into a new meta-architecture added the strengths of both models, and self-supervised learning helped learn meaningful representations without the need for explicit labels.

The production of solar energy at altitudes higher than 3800 m above sea level is not constant because the relevant factors are highly varied and complex due to extreme solar radiation, climatic variations, and hostile environments. Therefore, it is necessary to create efficient prediction models to forecast solar production even before implementing photovoltaic systems. In this study, stacking techniques using ElasticNet and XGBoost were applied in order to develop regression models that could collect a maximum number of features, using the LASSO, Ridge, ElasticNet, and Bayesian models as a base. A sequential feature selector (SFS) was used to reduce the computational cost and optimize the algorithm. The models were implemented with data from a string photovoltaic (PV) system in Puno, Peru, during April and August 2021, using 15 atmospheric and photovoltaic system variables in accordance with the European standard IEC 61724-20170. The results indicate that ElasticNet reduced the MAE by 30.15% compared to the base model, and that the XGBoost error was reduced by 30.16% using hyperparameter optimization through modified random forest research. It is concluded that the proposed models reduce the error of the prediction system, especially the stacking model using XGBoost with hyperparameter optimization.

The agro-industrial sector faces significant challenges in product classification, which directly affect product quality, production efficiency and food safety. This paper proposes a machine learning model that correctly identifies the different attributes of Persea americana. For this, an automatic agro-industrial plant was implemented following industrial standards where advanced image processing techniques were used on a dataset of 346 images for training and 146 images for testing, with three deep convolutional neural networks with improved training strategies and advanced validation techniques including True Skill Statistic (TSS), Cohen’s Kappa (K), Threat Score (TS), Heidke Skill Score (HSS) and Probability of Error (Pe). The results showed that the DenseNet model outperforms other state-of-the-art models in accuracy, reaching an F1 score of 99.27%, ResNet 50 reached 99.26% and EfficientNet B4 reached 99.19%, also in the validation phase TSS, K, TS and HSS for all models were higher than 0.98 while the Pe index was higher than 0.55. It is concluded that the DenseNet model is shown to be the effective and reliable technique for the classification of Persea Americana. These promising results open new possibilities for the implementation of machine learning in the agri-food industry. For future research, the possibility of expanding the data set and extending the application of this model to other varieties of fruits is proposed.

Global avocado production exceeds 10 million tons annually. Among the diseases affecting avocado fruit, anthracnose is one of the most significant, causing black lesions and fruit decay that can result in yield losses of 20–30%. To facilitate the early detection of anthracnose, this study proposes a computer vision-based approach. A dataset containing 2218 images of Fuerte avocados was first developed, comprising 1730 healthy samples and 488 anthracnose-infected samples after the labeling process. In the experimental phase, several convolutional neural network (CNN) models with varying depths (3, 4, 5, and 6 layers) were designed and evaluated. These models were subsequently integrated into different weighted ensemble configurations, where the best performance was achieved by the ensemble combining all four individual CNNs. The proposed weighted ensemble was compared against widely used state-of-the-art architectures, including VGG-16, ResNet-18, and MobileNetV2. Experimental results demonstrated the effectiveness of the proposed approach, achieving an F1-score of 0.9052, outperforming VGG-16 (0.8283), ResNet-18 (0.7328), and MobileNetV2 (0.7320).

This study proposes the SS-XceVNet meta-architecture, which integrates Xception and VGG16 to detect structural peeling in rigid pavements in an automated manner, thereby reducing the reliance on large, labeled datasets. With increasing traffic and exposure to environmental factors, failures in rigid pavement joints and corners are becoming increasingly frequent, making manual inspections time consuming, dangerous, and subjective. To address this problem, a method was developed in four phases. First, photogrammetry and data augmentation techniques are used to collect and increase the variety of images. Self-supervised learning was then applied to unlabeled data to reduce reliance on manual annotations. SS-XceVNet was then trained on labeled data, combining the robustness of Xception and VGG16 to refine the accuracy. Finally, the results were compared with those of the XceVNet-based model. Evaluations showed that SS-XceVNet achieved 80.43% accuracy, surpassing 47.83% of the base model. This underscores the usefulness of self-supervised learning and integration of advanced architectures in pavement failure detection. In conclusion, this approach improves efficiency in the functional evaluation of road infrastructures while reducing costs and occupational risks, laying the foundation for future research that contributes to the implementation of proactive maintenance systems. Automated detection thus strengthens the quality control of pavements, favoring road safety, and prolonging the useful life of roads.