South China has a climate characteristic of high temperature and high humidity, and the temperature and relative humidity inside a Venlo greenhouse are higher than those in the atmosphere. This paper studied the effect of ventilation conditions on the spatial and temporal distribution of temperature and relative humidity in a Venlo greenhouse. Two ventilation conditions, with and without a fan-pad system, were studied. A GA + BP neural network was applied to predict the temperature and relative humidity in fan-pad ventilation in the greenhouse. The results show that the temperature in the Venlo greenhouse ranged from 15.8 °C to 48.5 °C, and the relative humidity ranged from 24.9% to 100% during the tomato-planting cycle. The percentage of days when the temperature exceeded 35 °C was 67.3%, and the percentage of days when the average relative humidity exceeded 70% was 83.7%. The maximum temperature differences between the three heights under NV (Natural Ventilation) and FPV (Fan-pad Ventilation) conditions were 3.4 °C and 4.5 °C, respectively. The maximum relative humidity differences between the three heights under NV and FPV conditions were 8.4% and 21.7%, respectively. The maximum temperature difference in the longitudinal section under the FPV conditions was 3.2 °C, while the relative humidity was 11.4%. The cooling efficiency of the fan-pad system ranged from 16.6% to 70.2%. The non-uniform coefficients of the temperature under the FPV conditions were higher than those under the NV conditions, while the nonuniform coefficients of the relative humidity were the highest during the day. The R², MAE, MAPE and RMSE of the temperature-testing model were 0.91, 0.94, 0.11, and 1.33, respectively, while those of relative humidity model were 0.93, 2.83, 0.10, and 3.86, respectively. The results provide a reference for the design and management of Venlo greenhouses in South China.

Acquiring real-time feeding information for monitoring lactating sows and their feeding requirements is a challenging task. Real-time data represent an important input for numerous tasks, such as disease monitoring, nutritional regulation, and feeding modeling. However, concurrently monitoring large numbers of sows and processing the real-time information for modeling is challenging using existing platforms. In this paper, we describe the design and development of a system that monitors and processes sows’ feed and water consumption in real time. The system was custom-developed using open-source networking technologies. The system consists of three components: an electronic sow feeder connected to a central controller via a CAN network, an MQTT service cluster, and a data processing program. The MQTT service cluster uses Netty to develop a single service node, and it uses Zookeeper and Redis to complete node registration, discovery, and scheduling. The data processing program is based on Spark and Flink. We conducted comparative testing of three common codecs (Java Serializer, Marshalling, and Protostuff) to further speed up data transmission. The results of the experiment show that, with three service nodes, the system can concurrently monitor up to 20,000 sows. Moreover, the system achieves optimal performance when monitoring 10,000 sows at the same time, with a TPS of 6399 pcs/s and an RT of 643 ms.

Greenhouses are applied to mitigate the deleterious effects of inclement weather, which facilitates the optimal growth and development of the crops. South China has a climate characterized by high temperature and high humidity, and the temperature and relative humidity inside a Venlo greenhouse are higher than those in the atmosphere. In this paper, the numerical model of the flow distribution of a Venlo greenhouse in South China was established using the CFD method, which mainly applied the DO model, the k-e turbulence model, and the porous medium model. The porous resistance characteristics of tomatoes were obtained through experimental research. The inertial resistances of tomato plants in the x, y, and z directions were 80,000,000, 18,000,000, and 120,000,000, respectively; the viscous resistances of tomato plants in the x, y, and z directions were 0.43, 0.60, and 0.63, respectively. The porosity of tomato plants was 0.996. The average difference between the temperature of the established numerical model and the experimental temperature was less than 0.11 °C, and the average relative error was 2.72%. This research also studied the effects of five management and structure parameters on the velocity and temperature distribution in a greenhouse. The optimal inlet velocity is 1.32 m/s, with the COF of velocity and temperature being 9.23% and 1.18%, respectively. The optimal skylight opening is 1.76 m, with the COF of velocity and temperature being 10.68% and 0.88%, respectively. The optimal side window opening is 0.67 m, with the COF of velocity and temperature being 9.25% and 2.10%, respectively. The optimal side window height is 1.18 m, with the COF of velocity and temperature being 9.50% and 1.33%, respectively. The optimal planting interval is 1.40 m, with the COF of velocity and temperature being 15.29% and 0.20%, respectively. The results provide a reference for the design and management of Venlo greenhouses in South China.

Continuous sensing of group feeding demand is needed to support feeding decisions on large-scale pig farms. In dense group-housing environments, crowding, occlusion, and variation in posture around the trough can reduce the stability of visual feeding-status recognition. Feeding schedules also commonly rely on fixed periods or historical patterns, limiting their ability to assess whether daily feeding-behavior signals are aligned with group demand. Here, we propose a video-based workflow for sensing group feeding activity and conducting an offline replay evaluation of feeding-schedule allocation rules in group-housed pigs. Using synchronized feeding videos and feeder records, we developed a workflow that integrates rotated object detection, visual feeding intensity calculation, time-period-level construction of video-derived feeding demand, and feeding strategy evaluation. Rotated bounding-box detection achieved an mAP50-95 of 0.908 for feeding recognition, supporting its feasibility for feeding/non-feeding recognition in this dense-pen scene. Visual feeding intensity derived from model outputs was most strongly correlated with actual feed intake increment under a 20 min aggregation window with no time lag (r = 0.80). In the evaluation based on video-derived feeding demand, the video-perceived hybrid strategy reduced the peak time difference to 14.51 min, achieved a synchronization correlation of 0.611, and reduced the mismatch rate to 5.13%. These findings suggest that video-based behavioral signals can serve as a group-level proxy for feeding demand and can support offline evaluation of behavior-aware feeding schedules. However, the demand and strategy analyses were based on a single pen observed for 72 h, and validation across additional pens, batches, age stages, seasons, and closed-loop feeding conditions is required before practical deployment.