Aldo Calcante

The ready-to-eat lettuce industry is rapidly expanding, increasing the need for reliable, scalable methods to assess seed germination and early growth under realistic soil conditions. This study presents an automated imaging-based approach for quantifying germination dynamics and seedling vigor using a low-cost multi-camera system under greenhouse conditions. Lettuce seeds were grown in soil either inoculated or non-inoculated with the soil-borne pathogen Rhizoctonia solani. Top-view images were acquired using commercial surveillance cameras and processed through a calibrated pipeline including geometric correction, color normalization, vegetation segmentation, clustering, and temporal tracking of emergence events. Seedling vigor was quantified through projected leaf area estimation. The proposed method enables accurate estimation of germination kinetics and growth dynamics under field-like conditions. Automated counts were validated against manual measurements at both intermediate and final time points, achieving high agreement in both cases. At the final assessment, the method reached R² = 0.98 and RMSE = 1.12, while at the midterm evaluation it achieved improved performance with R² = 0.998 and RMSE = 0.5, reflecting the lower complexity of plant structure at earlier growth stages. Results showed that pathogen inoculation significantly reduced both germination rate and seedling vigor, with up to 70% reduction in biomass accumulation. The proposed framework provides a robust, low-cost solution for high-throughput phenotyping of early plant development in soil-based systems, supporting scalable agricultural experimentation.

Crop phenotyping refers to the systematic measurements of plant traits, including morphological and structural characteristics, such as plant size, leaf area, flower density, biomass growth, in response to environmental conditions, biotic and abiotic stress, and management treatments. Phenotyping is crucial in modern agriculture for developing improved crop varieties, increasing yield and resource-use efficiency. Traditional phenotyping approaches are highly labour-intensive and time-consuming, especially under field conditions, where manual measurements are difficult and often prone to inconsistencies and inaccuracies. Over the last decades, advances in imaging and sensor technology have progressively shifted research toward non-destructive, high-throughput phenotyping methods, with three-dimensional (3D) imaging recently emerging as a particularly promising tool.In this study, a traditional two-dimensional (2D) RGB imaging workflow was compared with a 3D RGB-D imaging approach, based on a low-cost Microsoft Kinect v2 sensor, for non-contact biomass estimation of lettuce plants measured from a top-view configuration.The objective was to assess whether 3D volumetric descriptors extracted from point-cloud data can provide better estimation accuracy and robustness than conventional 2D projected canopy area, particularly across different growth stages and canopy architectures. This evaluation is essential for developing reliable, low-cost, and scalable phenotyping solutions suitable for greenhouse automation, high-throughput cultivar screening, and digital field monitoring platforms.The 3D approach involved the development and evaluation of two algorithms. The first algorithm performs voxelization of the point cloud corresponding to each individual plant, while the second builds upon this step by incorporating a reconstruction technique to account for plant structures occluded from the top-view perspective, thereby expected to further enhancing the accuracy of biomass estimation. The 2D RGB-based method achieved an R2 of 0.88 (RMSE = 0.27 g; MAPE = 18.3%) in estimating the plant biomass, as determined by destructive harvesting and leaves dry matter weight. In contrast, the 3D approach significantly improved prediction accuracy, reaching an R2 of 0.94 (RMSE = 0.21 g; MAPE = 16.1%) with the voxelization method alone, and an R2 of 0.98 (RMSE = 0.11 g; MAPE = 11.6%) when combined with the reconstruction algorithm. These results demonstrate the superior robustness and reliability of low-cost 3D imaging system for biomass estimation, and highlight the critical role of the dedicated volumetric reconstruction algorithm proposed in this study. The findings confirm that, in addition to advanced sensing technologies, the design of tailored and high-performing algorithms is essential for accurate and reliable biomass assessments, supporting low-cost solutions for precision agriculture and plant breeding applications.

The optimal management of cattle nutrition promotes animal health and welfare, increases livestock farms’ productivity and competitiveness, and enhances environmental sustainability practices. Animal feeding operations play a crucial role as many factors can drive the theoretical ration formulated by nutritionists far from the one the animals ingest. Precision feeding technologies (e.g., NIR sensors on the milling cutter of the chopper-mixer wagon; computer vision systems installed in the mixing tank) may allow for accurate and real-time analysis of the chemical and physical properties of total mixed ration (TMR) ingredients, reducing errors during its preparation and distribution. This work compares the physical quality and the length of the fibre of the TMR resulting from the chopping-mixing process of a conventional mixing wagon, one machine-learning-assisted mixing wagon and an automatic feeding system under actual operating conditions. Between October 2021 and November 2022, TMR sampling occurred on four dairy farms and one fattening bulls farm in Northern Italy, specifically in the Brescia, Cremona, and Mantua districts. TMR samples underwent particle size analysis using the Penn State Particle Separator (PSPS) method and, once in the laboratory, moisture analysis and fibre length measurement. Concerning TMR particle size analysis, the PSPS method revealed that the machine learning-assisted mixing wagon provided TMR with physical features comparable to that from ordinarily run mixing wagons. At the same time, the automatic feeding system resulted in TMR with finer particle size, following the farmers’ choice not to use long-stemmed forages. Regarding fibre length, only the TMR resulting from the operator-based mixing wagon aligned with the targeted fibre length of 5 cm, while the AFS and the ML-assisted mixing resulted in higher fibre lengths. Overall, the use of computer vision (CV) systems is helpful for the consistency of the TMR and represents a valuable solution for animal farming, particularly when employing low- or inexperienced operators. Further studies are, however, needed to improve the training of the with elements that can replicate the operator experience.

Early diagnosis of subclinical ketosis is fundamental in the production management of dairy cattle. Without evident clinical signs, this pathological condition causes important economic losses for the farmer and significant health repercussions for the cattle that could develop an altered immune function. Laboratory techniques, although accurate, are expensive, invasive, and cannot be used for real-time monitoring of the entire herd. On the contrary, the analysis of volatile organic compounds (VOCs) contained in the breath of dairy cattle affected by ketosis could represent a key biomarker of the ketogenic process. For this reason, we developed a sensory device, tested in the laboratory, to detect acetone concentrations ranging from 1 to 10 ppm (concentrations typically detected in the cow’s breath), and we look to verify the electronic nose’s potential as a non-invasive diagnostic tool for ketosis. Experimental results show the high sensitivity of the instrument in differentiating acetone solutions. Principal component analysis (PCA) showed a clear separation of samples in the score plot, while classification using linear discriminant analysis (LDA) and quadratic discriminant analysis (QDA) achieved accuracy rates above 70% and 85%, respectively. These findings suggest the potential application of the electronic nose as a non-invasive diagnostic tool in veterinary diagnostic studies. In particular, its ability to detect and discriminate low acetone concentrations could help the farmer to improve the overall management of the herd, optimising monitoring strategies and ketosis diagnosis before the appearance of the clinical signs of the disease.

The expansion of dairy cattle farms and the increase in herd size have made the control and management of animals more complex, with potentially negative effects on animal welfare, health, productive/reproductive performance and consequently farm income. Precision Livestock Farming (PLF) is based on the use of sensors to monitor individual animals in real time, enabling farmers to manage their herds more efficiently and optimise their performance. The integration of sensors and devices used in PLF with the Internet of Things (IoT) technologies (edge computing, cloud computing, and machine learning) creates a network of connected objects that improve the management of individual animals through data-driven decision-making processes. This paper illustrates the main PLF technologies used in the dairy cattle sector, highlighting how the integration of sensors and devices with IoT addresses the challenges of modern dairy cattle farming, leading to improved farm management.