Mastitis is a significant challenge in the buffalo industry, affecting both milk production and animal health and resulting in economic losses. This study presents the first fully automated AI-driven thermal imaging system integrated with robotic milking, specifically developed for the real-time, non-invasive monitoring of udder health in Italian Mediterranean buffalo. Unlike traditional approaches, the system leverages the synchronized acquisition of thermal images during milking and compensates for environmental variables through a calibrated weather station. A transformer-based neural network (SegFormer) segments the udder area, enabling the extraction of maximum udder skin surface temperature (USST), which is significantly correlated with somatic cell count (SCC). Initial trials demonstrate the feasibility of this approach in operational farm environments, paving the way for scalable, precision diagnostics of subclinical mastitis. This work represents a critical step toward intelligent, automated systems for early detection and intervention, improving animal welfare and reducing antibiotic use.
1 September 2019 in International Immunopharmacology
A three‐year‐old entire male French bulldog was brought in with the following clinical signs: apathy, vomiting and abdominal distension. Physical examination showed dehydration, tachypnoea and presence of jaundice. Abdominal radiology revealed the presence of an important loss of details in the abdominal structures compatible with peritoneal effusion, confirmed by abdominal ultrasonography. Findings detected by ultrasonography consisted of splenomegaly with alteration of the echotexture, hepatomegaly with presence of nodules and irregular areas in the liver capsule. An ultrasound‐guided sample of the peritoneal fluid was taken for laboratory analysis. Abnormal clinicopathological results included pancytopenia, hyperglobulinaemia, hypoalbuminaemia, and elevation of the hepatic parameters and total bilirubin. High antibodies to Leishmania infantum were detected by ELISA in serum and peritoneal fluid samples. However, PCR for L infantum showed a negative result from the peritoneal fluid sample. This is the first reported case of canine leishmaniosis with the specific detection of L infantum antibodies in peritoneal effusion.
The growing population of wild boars (Sus scrofa) in all highly anthropized countries represents a current challenge for the protection of ecosystems, agriculture and urban environments. This study introduces an innovative capture solution based on IoT systems designed to enable the selective capture of sub-adult wild boars in an ethical manner and minimize stress and bycatch. Conducted over five years in a Natura 2000 area in Campania, Italy, the research integrates advanced technologies, including AI-based infrared cameras, LoRa communication and autonomous feeding systems, to monitor, control and operate a specially designed selective cage trap. The results obtained demonstrate how technological innovation improves wildlife and hunting management by selecting younger animals without interfering with group dynamics. Selective capture ensures healthy population control, does not conflict with hunting and reduces pressure on habitats, especially if these fall within areas of particular importance for European biodiversity.
Accurate and automatic extraction of the udder skin surface temperature (USST) from thermal imagery enables continuous, non-invasive monitoring of udder health. The literature indicates that USST correlates with somatic cell count (SCC), the primary marker of mastitis activity, provided that udder regions are precisely segmented. Building on our previous system deployed on-farm, which integrated thermal imaging with robotic milking for real-time monitoring, we focus this paper on the segmentation component to obtain a precise yet lightweight, farm-deployable model. We explore compact binary segmentation models that pair the newly released DINOv3-initialised encoders with lightweight decoders: ConvNeXt + Feature Pyramid Network (FPN) and ViT + Dense Prediction Transformer (DPT). Our training follows four stages: (i) train the decoder while keeping the encoder frozen; (ii) apply isomorphic pruning to remove redundant, shape-consistent channels and feature dimensions while preserving tensor interfaces; (iii) unfreeze the pruned encoder and fine-tune the whole network to recover performance; and (iv) apply post-training quantisation to produce FP16 variants. We evaluate encoders across scales and report test intersection over union (IoU), parameter count, and multiply–accumulate operations (MACs), selecting via the IoU-cost frontier at full precision. While a non-pruned ConvNeXttiny backbone attains the highest IoU, a pruned ConvNeXtsmall achieves comparable IoU at a lower cost and is therefore our Pareto-optimal choice for edge deployment. Such a model (ConvNeXtsmall+FPN) achieves an 81.68% IoU with 25.44M parameters and 32.60B MACs. It is exported to ONNX and quantised, and deployed over an Nvidia Jetson Orin Nano as a practical, offline, and privacy-preserving edge solution for farms. The deployed FP16-quantised ONNX model with CUDA-accelerated runtime achieves an average latency of 81.6 ms per frame at the 15W power profile on the Jetson Orin Nano. The dataset, source code, training logs, model weights, and Jetson-ready Docker image used for this study are released open-source to ensure full reproducibility and a fair comparison.