Early detection of tomato bacterial spot is critical for effective disease management in greenhouse environments. This study developed an automated hyperspectral data collection system to capture spectral information from inoculated tomatoes in a walk-in growth chamber, enabling noninvasive monitoring of disease progression. A plant physiological-informed machine learning model, incorporating a genetic algorithm for feature selection, was used to optimize the classification of healthy and diseased samples. To reduce multicollinearity and enhance the machine learning model performance, three key vegetation indices (VIs) were selected from a pool of 18 candidate VIs using a genetic algorithm. The model achieved a 92% recall rate using the three VIs that are associated with chlorophyll content and photosynthetic efficiency. The result of this study suggests a promising methodology for plant disease detection beyond that of the tomato bacterial spot.

This project aimed to enhance local weather forecasts by improving 1-hour, on-site predictions using the High-Resolution Rapid Refresh (HRRR) dataset. These forecasts can support the high tunnel weather forecast model, providing growers with critical insights to respond to high-temperature events. The project’s objectives included developing a streamlined data preparation process and an on-site predictive model using machine learning (ML). After considering all potential weather variables, our analysis focused on solar radiation intensities exceeding 400 W/m 2 during the Northern Hemisphere’s transition periods (March and October). The study used HRRR and observational data from three locations, including Wooster, OH, USA; West Lafayette, IN, USA; and Geneva, NY, USA for model training. Data preprocessing, including parsing, time synchronization, format unification, and missing data handling, was managed using Python. The complex meteorological HRRR data, originally in GRIB2 format, was transformed into a more accessible CSV format with selected variables and a significantly reduced file size, making it more usable for high tunnel producers. For the ML model, one neural network architecture effectively served all three locations, suggesting the potential for a generalized model that can be applied across sites at similar latitudes. Among the five input-feature designs, the HRRR forecast variables for the current time and next hour performed the best across all locations. The ML model outperformed HRRR, reducing root mean square error (RMSE) from 114 to 64 W/m 2 and mean error from 34 to 4 W/m 2 while improving R 2 from 0.47 to 0.67 for Wooster, OH. Similar performance gains were observed at the other locations. These findings support broader agricultural applications, including high tunnels, greenhouses, and outdoor farming.

Kale (Brassica napus) and collard (Brassica oleracea) are two leafy greens in the family Brassicaceae. The leaves are rich sources of numerous health-beneficial compounds and are commonly used either fresh or cooked. This study aimed to optimize the nutrient management of kale and collard in hydroponic production for greater yield and crop quality. ‘Red Russian’ kale and ‘Flash F1’ collard were grown for 4 weeks after transplanting in a double polyethylene-plastic-covered greenhouse using a nutrient film technique (NFT) system with 18 channels. Kale and collard were alternately grown in each channel at four different electrical conductivity (EC) levels (1.2, 1.5, 1.8, and 2.1 mS·cm−1). Fresh and dry yields of kale increased linearly with increasing EC levels, while those of collard did not increase when EC was higher than 1.8 mS·cm−1. Kale leaves had significantly higher P, K, Mn, Zn, Cu, and B than the collard at all EC levels. Additionally, mineral nutrients (except N and Zn) in leaf tissue were highest at EC 1.5 and EC 1.8 in both the kale and collard. However, the changing trend of the total N and NO3- of the leaves showed a linear trend; these levels were highest under EC 2.1, followed by EC 1.8 and EC 1.5. EC levels also affected phytochemical accumulation in leaf tissue. In general, the kale leaves had significantly higher total anthocyanin, vitamin C, phenolic compounds, and glucosinolates but lower total chlorophylls and carotenoids than the collard. In addition, although EC levels affected neither the total chlorophyll or carotenoid content in kale nor glucosinolate content in either kale or collard, other important health-beneficial compounds (especially vitamin C, anthocyanin, and phenolic compounds) in kale and collard leaves reduced with the increasing EC levels. In conclusion, the kale leaf had more nutritional and phytochemical compounds than the collard. An EC level of 1.8 mS·cm−1 was the optimum EC level for the collard, while the kale yielded more at 2.1 mS·cm−1. Further investigations are needed to optimize nitrogen nutrition for hydroponically grown kale.

Most prior semantic segmentation methods have been developed for day-time scenes, while typically underperforming in night-time scenes due to insufficient and complicated lighting conditions. In this work, we tackle this challenge by proposing a novel night-time semantic segmentation paradigm, i.e., disentangle then parse (DTP). DTP explicitly disentangles night-time images into light-invariant reflectance and light-specific illumination components and then recognizes semantics based on their adaptive fusion. Concretely, the proposed DTP comprises two key components: 1) Instead of processing lighting-entangled features as in prior works, our Semantic-Oriented Disentanglement (SOD) framework enables the extraction of reflectance component without being impeded by lighting, allowing the network to consistently recognize the semantics under cover of varying and complicated lighting conditions. 2) Based on the observation that the illumination component can serve as a cue for some semantically confused regions, we further introduce an Illumination-Aware Parser (IAParser) to explicitly learn the correlation between semantics and lighting, and aggregate the illumination features to yield more precise predictions. Extensive experiments on the night-time segmentation task with various settings demonstrate that DTP significantly outperforms state-of-the-art methods. Furthermore, with negligible additional parameters, DTP can be directly used to benefit existing day-time methods for night-time segmentation.

To serve the intricate and varied demands of image editing, precise and flexible manipulation of image content is indispensable. Recently, DragGAN has achieved impressive editing results through point-based manipulation. However, we have observed that DragGAN struggles with miss tracking, where DragGAN encounters difficulty in effectively tracking the desired handle points, and ambiguous tracking, where the tracked points are situated within other regions that bear resemblance to the handle points. To deal with the above issues, we propose FreeDrag, which adopts a feature-oriented approach to free the burden on point tracking within the point-oriented methodology of DragGAN. The FreeDrag incorporates adaptive template features, line search, and fuzzy localization techniques to perform stable and efficient point-based image editing. Extensive experiments demonstrate that our method is superior to the DragGAN and enables stable point-based editing in challenging scenarios with similar structures, fine details, or under multi-point targets.

An electronic nose (E-nose) system equipped with a gas sensor array and real-time control panel was developed for a fast diagnosis of whitefly infestation in tomato plants. Profile changes of volatile organic compounds (VOCs) released from tomato plants under different treatments (i.e., whitefly infestation, mechanical damage, and no treatment) were successfully determined by the developed E-nose system. A rapid sensor response with high sensitivity towards whitefly-infested tomato plants was observed in the E-nose system. Results of principal component analysis (PCA) and hierarchical clustering analysis (HCA) indicated that the E-nose system was able to provide accurate distinguishment between whitefly-infested plants and healthy plants, with the first three principal components (PCs) accounting for 87.4% of the classification. To reveal the mechanism of whitefly infestation in tomato plants, VOC profiles of whitefly-infested plants and mechanically damaged plants were investigated by using the E-nose system and GC-MS. VOCs of 2-nonanol, oxime-, methoxy-phenyl, and n-hexadecanoic acid were only detected in whitefly-infested plants, while compounds of dodecane and 4,6-dimethyl were only found in mechanically damaged plant samples. Those unique VOC profiles of different tomato plant groups could be considered as bio-markers for diagnosing different damages. Moreover, the E-nose system was demonstrated to have the capability to differentiate whitefly-infested plants and mechanically damaged plants. The relationship between sensor performance and VOC profiles confirmed that the developed E-nose system could be used as a fast and smart device to detect whitefly infestation in greenhouse cultivation.

HighlightsAn algorithm was developed to process laser sensor data to make more accurate measurements of canopy dimensions.The algorithm isolated individual canopies, removed distortion, and estimated the occluded portions of the dataset.The algorithm reduced measured error by 46% in terms of root mean square error (RMSE).The RMSE was higher for sensor heights below and above a calculated optimal sensor height.Abstract. Laser-guided intelligent spray technology for greenhouse applications requires sensors that can accurately measure plant dimensions. This study proposed a new method to overcome current limitations by introducing a processing algorithm that manipulates the noisy dataset and determines the optimal sensor height to produce better measurements of the canopy width. The processing algorithm involves a combination of registration, clustering, and mirroring. Registration aligns multiple scans of the same scene to improve resolution. Clustering isolates individual plant canopies from the dataset to enable further processing. Mirroring is used to resolve the problems of distortion and occlusion and predict missing information in the dataset. The performance of the processing algorithm was evaluated by calculating the root mean square error (RMSE) in the canopy width measurements. Its results were compared with the measurements reported in earlier research, where there was limited processing of the laser sensor data. The processing algorithm reduced RMSE values by 46% compared to the earlier research, and the largest improvements were seen for objects placed beyond 1.5 m from the sensor. The sensor height was observed to be inversely proportional to the RMSE values. The average RMSE of the processing algorithm was 25 mm, compared to 47 mm in the earlier research when the laser sensor was at a height of 1 m. Another experimental setup was used to test the limits of the relationship between sensor height and algorithm performance while using objects that were more representative of plant canopy shapes. The accuracy of the processing algorithm decreased when the sensor height was either above or below the optimal sensor height, which was derived from calculations made in earlier research. The processing algorithm has potential to improve spray efficiencies. Keywords: Automation, Clustering, LiDAR, Point cloud data processing, Variable-rate spray.

Arugula (Eruca sativa) is cultivated using hydroponic techniques in greenhouses to fulfill high year-round demand, but its nutrient management in hydroponic production has not yet been standardized, potentially leading to limited quality and productivity. Aiming to address this issue, we investigated the effect of electrical conductivity (EC) on yield, nutritional and phytochemical properties of arugula. The model cultivar arugula ‘Standard’ was grown at four different EC levels (1.2, 1.5, 1.8, and 2.1 dS·m−1). Our results indicated photosynthetic properties, SPAD, leaf area, yield and dry weight increased with increasing EC from 1.2 to 1.8 dS·m−1. Foliar nutrient content increased with higher EC, but nutrient solution with 2.1 dS·m−1 showed a significant decline in N, Ca and most of the micronutrients including Fe, Zn, Mo, Cu, B and Mn. Total glucosinolates, total chlorophyll and total carotenoids concentrations increased with increasing EC. In addition, total anthocyanin content was highest in plants grown in EC 1.2 and 2.1 dS·m−1, demonstrating a stress response when grown in extreme EC levels. Our results further indicated a rapid accumulation of nitrate with higher EC, potentially detrimental to human health. This research demonstrated the optimal EC range would be 1.5 to 1.8 dS·m−1 for arugula in hydroponic production systems based on yield, quality criteria and human health considerations.

In greenhouse energy balance models, the soil thermal parameters are important for evaluating the heat transfer between the greenhouse air and the soil. In this study, the soil thermal diffusivity was estimated from greenhouse soil temperature data using the amplitude, phase-shift, arctangent, logarithmic, and min-max methods. The results showed that the amplitude method and the min-max method performed well in estimating the soil thermal diffusivity. The obtained soil thermal diffusivity was input into a sinusoidal model to determine the greenhouse soil temperature at different soil depths. For greenhouse applications, the daily average soil temperature at different depths was predicted according to the temperature at the surface and the annual mean soil temperature. The model was validated using soil temperature data from summer and winter, when the greenhouse was cooled and heated, respectively.

An electronic nose (E-nose) system equipped with a sensitive sensor array was developed for fast diagnosis of aphid infestation on greenhouse tomato plants at early stages. Volatile organic compounds (VOCs) emitted by tomato plants with and without aphid attacks were detected using both the developed E-nose system and gas chromatography mass spectrometry (GC-MS), respectively. Sensor performance, with fast sensor responses and high sensitivity, were observed using the E-nose system. A principle component analysis (PCA) indicated accurate diagnosis of aphid-stressed plants compared to healthy ones, with the first two PCs accounting for 86.7% of the classification. The changes in VOCs profiles of the healthy and infested tomato plants were quantitatively determined by GC-MS. Results indicated that a group of new VOCs biomarkers (linalool, carveol, and nonane (2,2,4,4,6,8,8-heptamethyl-)) played a role in providing information on the infestation on the tomato plants. More importantly, the variation in the concentration of sesquiterpene VOCs (e.g., caryophyllene) and new terpene alcohol compounds was closely associated with the sensor responses during E-nose testing, which verified the reliability and accuracy of the developed E-nose system. Tomato plants growing in spring had similar VOCs profiles as those of winter plants, except several terpenes released from spring plants that had a slightly higher intensity.

Improving the coverage area and fading time of herbicidal droplets on weeds has the potential to enhance the biological control effectiveness. Droplet spreading and fading behaviors on five different weeds were characterized for spray solutions containing a 1.25% glyphosphate Rodeo herbicide amended with each of three different adjuvants (nonionic surfactant Kinetic, nonionic organosilicone surfactant DyneAmic, and nonionic surfactant and antifoaming agent Preference). The five weeds were ragweed, crabgrass, yellow nutsedge, common purslane, and spurge. Tests were conducted by depositing single 300 and 600 mu m herbicidal droplets with different adjuvant concentrations on weed leaves inside an environment control chamber. A droplet at a higher adjuvant concentration had greater coverage area on weed surfaces. Preference-amended herbicidal droplets had the largest coverage area increase for all five weeds, and generally followed by droplets with Kinetic and DyneAmic except for 300 mu m droplet on purslane and 600 mu m droplet on spurge. In comparison with the herbicidal solution containing Rodeo and water only, with addition of adjuvants the 600 mu m droplets increased the coverage area by 2.13 to 5.47, 1.76 to 2.56, 1.84 to 2.07, and 2.40 to 4.49-fold on crabgrass, yellow nutsedge, common purslane, and spurge, respectively, while the 300 mu m droplets increased the coverage area on ragweed by 3.88 to 5.86-fold. In contrast, fading times of all 300 mu m droplets decreased with the adjuvant addition except for DyneAmic applied on purslane. However, fading times of 600 mu m droplets did not have increase or decrease trends with adjuvants, which depended on types of the adjuvant and weed. The overall comparison by integrated index (coverage area x fading time) indicated that a spray droplet at higher adjuvant concentration was likely to have a higher integrated index. In addition, Preference amended droplets had significantly more integrated index increase for crabgrass and nutsedge, while DyneAmic had more increase for purslane. Therefore, appropriate selections of spray adjuvants during herbicide applications could significantly increase droplet deposition effectiveness for controlling specific weeds.

Nutritional recommendations for hydroponic crops in nutrient film techniques (NFT) are given based on electrical conductivity (EC) levels of the fertilizer solution. Growers rely on the recommended EC levels of 1.8 mS cm-1 for lettuce production, despite how growing conditions vary in the greenhouses throughout the year. Disorders like tipburn are a common occurrence in hydroponic lettuce due to this overreliance on EC. Variation of yield with different EC levels were reported in hydroponic systems, however, reliable data are not available for the interactions between EC, nutrient uptake patterns and yield in NFT systems under greenhouse conditions in northern USA. Therefore, two experiments were conducted in fall 2016 and spring 2017, with lettuce cultivars grown under four EC levels, 0.8, 1.2, 1.8 and 2.4 mS cm-1. The temperature set points for fall and spring trials were 65 to 70 degrees F/18.3 to 21.1 degrees C and 70 to 75 degrees F/21.1 to 32.8 degrees C, respectively. During the fall trial, maximum yield was observed at EC of 1.8 mS cm(-1) with 225 +/- 9 g per head in `Green Butter' lettuce and 213 +/- 13 g per head in `Red Butter' lettuce. Nitrogen and potassium uptake by the leaves increased with increased EC, however, there was no influence on yield above 1.8 mS cm(-1) EC, since both fresh and dry weight were reduced with EC greater than 1.8 mS cm(-1) in both cultivars. Phosphorus, Ca, Mg, S, Mo and Mn content of leaves increased from EC of 1.2 to 1.8, however, remained similar between 1.8 and 2.4 mS cm(-1). Photosynthetic rate increased with increasing EC and leveled off after 1.8 mS cm(-1). Tipburn severity is greatest at 1.8 mS cm-1 EC. During spring trial there were no yield differences between the 1.8 and 2.4 mS cm-1 EC. Increasing EC beyond 1.8 mS cm(-1) did not contribute to yield gains, however, tipburn symptoms were greatest at EC at 1.8 mS cm(-1).

Precision variable-rate spraying technology is needed for controlled-environment plant production in greenhouses. An experimental spray system for greenhouse applications was developed for real-time control of individual nozzle outputs. The system mainly consisted of a high-speed laser scanning sensor, 12 individual variable-rate nozzles, an embedded computer, a spray control unit, and a 3.6 m long mobile spray boom. Each nozzle was coupled with a pulse-width modulated solenoid valve to discharge sprays at variable rates based on target presence and plant canopy structure. Laboratory tests were conducted to evaluate the accuracy of the spray control system in respect to spray delay time, nozzle activation, and spray volume using four target objects of different regular geometrical shapes and surface textures and two artificial plants of different canopy structures. Other experimental variables included three detection heights from 0.5 to 1.0 m and five sensor travel speeds from 1.6 to 4.8 km h(-1). A high-speed video camera was used to determine the delay time and nozzle activation in discharging sprays on target objects after the laser sensor had detected the objects. The detection height and travel speed were found to have slight influence on the timing of nozzle activation. The nozzles started spraying in a range between 33 and 83 mm before reaching the target objects and stopped spraying between 13 and 84 mm after passing the objects, ensuring that the objects were fully covered by the spray. Spray volume corresponded to the object sizes well, and the spray control system performed with higher accuracy at lower travel speeds. Differences between the calculated spray volume based on the sensor detection and the actual spray volume ranged from 1.9 to 2.7 mL per object among all tested objects. The variable-rate control system reduced spray volume by 29.3% to 51.4% for all the objects compared with conventional constant-rate spraying. At the same time, the nozzles could be activated precisely by the object presence. Consequently, this experimental laser-guided system was implemented on a boom system in a commercial greenhouse for future investigations of its accuracy in variable-rate spraying to save pesticides, water, and nutrients.

Yield reduction resulting from high temperatures and tipburn are common issues during the summer for hydroponically grown lettuce using the nutrient–film technique (NFT). We investigated the yield and degree of tipburn of lettuce ‘Red Butter’, ‘Green Butter’, and ‘Red Oakleaf’ of the Salanova® series under different-solution electrical conductivity (EC) and pH levels. We also quantified the effect of foliar spray application of calcium chloride (CaCl2) on the yield and degree of tipburn using the lettuce cultivar Green Butter. For the EC experiment, the plants were grown at four EC levels (1.4, 1.6, 1.8, or 2.0 mS·cm–1) and a constant pH of 5.8. For the pH experiment, the plants were grown at and four pH levels (5.8, 6.0, 6.2, or 6.4) and a constant EC of 1.8 mS·cm–1. For the foliar spray experiment, CaCl2 was applied 1 week after transplanting into NFT channels at three different concentrations: 0, 200, 400, or 800 mg·L calcium (Ca). During the EC trial, the maximum yields were observed at or more than 1.8 mS·cm–1 for ‘Green Butter’ (263 ± 14 g/head) and ‘Red Butter’ (202 ± 8 g), and more than 1.6 mS·cm–1 for ‘Red Oakleaf’ (183 ± 6 g). The yield of ‘Green Butter’ was 75 g less at 1.4 mS·cm–1 compared with 1.8 mS·cm–1. Tipburn symptoms were less at 1.4 mS·cm–1 for ‘Green Butter’ whereas other cultivars were not highly susceptible. In pH trials, the maximum yield for all cultivars was found at pH 6.0 and 6.2. There were no differences in tipburn symptoms among all pH levels. The foliar spray treatment, twice a week at 400 or 800 mg·L–1 Ca, provided improved tipburn control, as the tipburn symptoms were minimal and the impact on yield was minor compared with reducing EC. This series of experiments found evidence in proper EC and pH management for optimum yield and tipburn control in NFT lettuce grown in summer conditions.

This paper reviews artificial intelligent noses (or electronic noses) as a fast and noninvasive approach for the diagnosis of insects and diseases that attack vegetables and fruit trees. The particular focus is on bacterial, fungal, and viral infections, and insect damage. Volatile organic compounds (VOCs) emitted from plants, which provide functional information about the plant’s growth, defense, and health status, allow for the possibility of using noninvasive detection to monitor plants status. Electronic noses are comprised of a sensor array, signal conditioning circuit, and pattern recognition algorithms. Compared with traditional gas chromatography–mass spectrometry (GC-MS) techniques, electronic noses are noninvasive and can be a rapid, cost-effective option for several applications. However, using electronic noses for plant pest diagnosis is still in its early stages, and there are challenges regarding sensor performance, sampling and detection in open areas, and scaling up measurements. This review paper introduces each element of electronic nose systems, especially commonly used sensors and pattern recognition methods, along with their advantages and limitations. It includes a comprehensive comparison and summary of applications, possible challenges, and potential improvements of electronic nose systems for different plant pest diagnoses.

A portable electronic nose (E-nose) system equipped with a sensitive sensor array was successfully developed to detect insect-stressed tomato plants, which were infested by aphids and whiteflies for 2 to 3 days, by taking advantages of their unique volatile organic compounds (VOCs) profiles. With showing fast sensor responses, an accurate diagnosis of aphids-stressed, whiteflies-stressed tomato plants from healthy groups were verified with PCA results accounting for 86% classification, which confirmed the promising capability of E-nose system providing a fast and reliable detection of infested tomato plants at early stage. For comparison, the changes of VOCs profiles were quantitatively determined by gas chromatography-mass spectrometry (GC-MS), with results showing that a group of new VOCs biomarkers (methyl salicylate and several terpenes) played info-chemical roles in the tomato-aphids interaction and tomato-whiteflies interaction, respectively. Moreover, the variation of the concentration of VOCs compounds explained the sensors behaviors during E-nose test, which confirmed the reliability and accuracy of the developed E-nose system. The satisfactory diagnosis among unstressed and insect stressed tomato plants as well as samples between, demonstrated the E-nose system had promising potential for a smart insect control at early stages in greenhouse.