Convolutional Neural Network Regression Research Articles

Prediction of a Cephalometric Parameter and Skeletal Patterns from Lateral Profile Photographs: A Retrospective Comparative Analysis of Regression Convolutional Neural Networks

Background/Objectives: Cephalometric analysis has a pivotal role in the quantification of the craniofacial skeletal complex, facilitating the diagnosis and management of dental malocclusions and underlying skeletal discrepancies. This study aimed to develop a deep learning system that predicts a cephalometric skeletal parameter directly from lateral profile photographs, potentially serving as a preliminary resource to motivate patients towards orthodontic treatment. Methods: ANB angle values and corresponding lateral profile photographs were obtained from the medical records of 1600 subjects (1039 female and 561 male, age range 3 years 8 months to 69 years 1 month). The lateral profile photographs were randomly divided into a training dataset (1250 images) and a test dataset (350 images). Seven regression convolutional neural network (CNN) models were trained on the lateral profile photographs and measured ANB angles. The performance of the models was assessed using the coefficient of determination (R2) and mean absolute error (MAE). Results: The R2 values of the seven CNN models ranged from 0.69 to 0.73, and the MAE values ranged from 1.46 to 1.53. Among the seven models, InceptionResNetV2 showed the highest success rate for predictions of ANB angle within 1° of range and the highest performance in skeletal class prediction, with macro-averaged accuracy, precision, recall, and F1 scores of 73.1%, 78.5%, 71.1%, and 73.0%, respectively. Conclusions: The proposed deep CNN models demonstrated the ability to predict a cephalometric skeletal parameter directly from lateral profile photographs, with 71% of predictions being within 2° of accuracy. This level of accuracy suggests potential clinical utility, particularly as a non-invasive preliminary screening tool. The system’s ability to provide reasonably accurate predictions without radiation exposure could be especially beneficial for initial patient assessments and may enhance efficiency in orthodontic workflows.

Analyzing convolutional neural networks and linear regression models for wind speed forecasting in sustainable energy provision

Wind energy's significance lies in its contribution to electricity production. Accurate wind speed prediction is crucial for precisely forecasting electricity generation, enhancing its overall importance. For the wind speed prediction two methods are developed in this paper that are the Linear Regression Method and Convolutional Neural Networks (CNNs). Linear regression makes a linear relation between input and output which are used to predict continuous outcomes. The second proposed leverage hierarchical feature extraction through convolutional layers, enabling them to excel at tasks like pattern recognition by capturing spatial patterns and hierarchies of information. Hyperparameter tuning is applied on both the models to minimize the errors. Then both the model’s performances are compared under different error matrices including MAE, RMSE, MSE and MAPE. The observations indicate that both the Linear Regression Model and CNN models are capable of forecasting wind speed However, the Linear Regression Model after Hyperparameter tuning performs better in terms of the calculated errors.

Rapid Detection of Protein and Starch Content in Brewing Wheat Using Hyperspectral Imaging Technology Combined with a Convolutional Neural Network Regression Model

Wheat is the main raw material in liquor brewing. However, the protein content (PC) and starch content (SC) in wheat will affect the quality and flavor of the final liquor. In this study, the rapid, non-destructive determination of the PC and SC in wheat was achieved by combining hyperspectral imaging (HSI) with a convolutional neural network regression (CNNR) model established using the original spectral data in the hyperspectral images. The best preprocessing method was first determined, and then the performance of CNNR, extreme gradient boosting (XGBoost) and partial least squares regression (PLSR) models were compared at full wavelength, and it was concluded that CNNR based on the original spectrum was the optimal model for predicting wheat PC (R-square (R2) = 0.9942, root mean square error (RMSE) = 0.1041, relative percentage difference (RPD) = 13.1306) and SC (R2 = 0.9329, RMSE = 0.8633, RPD = 3.8605) at full wavelength. Finally, to further explore the feature extraction capability of the CNN model, different feature selection and extraction methods are used to build the PLSR model, and the comparison revealed that the PLSR model built by feature extraction using convolutional neural network (CNN_F) performed best in predicting wheat PC and SC. These results showed that HSI combined with a CNNR model could enable the rapid and accurate analysis of the PC and SC in wheat, since the CNNR can extract features from samples better than the traditional machine learning models.

Regression convolutional neural network models implicate peripheral immune regulatory variants in the predisposition to Alzheimer's disease.

Alzheimer's disease (AD) involves aggregation of amyloid β and tau, neuron loss, cognitive decline, and neuroinflammatory responses. Both resident microglia and peripheral immune cells have been associated with the immune component of AD. However, the relative contribution of resident and peripheral immune cell types to AD predisposition has not been thoroughly explored due to their similarity in gene expression and function. To study the effects of AD-associated variants on cis-regulatory elements, we train convolutional neural network (CNN) regression models that link genome sequence to cell type-specific levels of open chromatin, a proxy for regulatory element activity. We then use in silico mutagenesis of regulatory sequences to predict the relative impact of candidate variants across these cell types. We develop and apply criteria for evaluating our models and refine our models using massively parallel reporter assay (MPRA) data. Our models identify multiple AD-associated variants with a greater predicted impact in peripheral cells relative to microglia or neurons. Our results support their use as models to study the effects of AD-associated variants and even suggest that peripheral immune cells themselves may mediate a component of AD predisposition. We make our library of CNN models and predictions available as a resource for the community to study immune and neurological disorders.

Advancing soil property prediction with encoder-decoder structures integrating traditional deep learning methods in Vis-NIR spectroscopy

The technology for estimating soil properties using visible and near-infrared spectroscopy has been maturing, with corresponding advances and breakthroughs in deep learning models. In this study, based on the large soil spectral library LUCAS, we explore the potential of encoder-decoder structures to improve convolutional neural network regression predictions. By introducing an encoder-decoder structure into the feature channels of a six-layer CNN model (TRNN model), we significantly enhanced the performance of shallow CNN models and successfully carried out regression predictions for seven soil properties. We employed IntegratedGradients, DeepLift, GradientShap, and DeepLiftShap methods to interpret the output of the TRNN model. Our TRNN model, built on raw spectra, demonstrated high accuracy in predicting multiple soil properties, outperforming residual architectures, LSTMs, various CNN architectures, and other traditional machine learning methods proposed in previous studies. We also investigated the impact of multi-task output structures (TRNN 1-M and TRNN M−M) and single-task output structures (TRNN 1-1) on model performance. For the TRNN model with an encoder-decoder structure, multi-task output structures resulted in a reduction in performance. The TRNN showed outstanding results in regression analysis of the seven soil properties selected in this study (cation exchange capacity, organic carbon content, calcium carbonate content, pH, clay content, silt content, and sand content), with R2 values exceeding 0.93 for all seven properties. Different soil characteristics correspond to different wavelengths, with multiple characteristic peaks commonly observed. This research convincingly demonstrates the enormous potential of combining large model architectures with traditional deep learning approaches for predicting soil properties, which could significantly advance precision agriculture.

Structural integrity of aging steel bridges by 3D laser scanning and convolutional neural networks

For steel bridges, corrosion has historically led to bridge failures, resulting in fatalities and injuries. To enhance public safety and prevent such incidents, authorities mandate in-situ evaluation and reporting of corroded members. The current inspection and evaluation protocol is characterized by intense labor, traffic delays, and poor capacity predictions. Here we combine full-scale experimental testing of a decommissioned girder, 3D laser scanning, and convolutional neural networks (CNNs) to introduce a continuous inspection and evaluation framework. Classification and regression CNNs are trained on a databank of 1,421 naturally inspired corrosion scenarios, generated computationally based on point clouds of three corroded girders collected in lab conditions. Results indicate low errors of up to 2.0% and 3.3%, respectively. The methodology is validated on eight real corroded ends and implemented for the evaluation of an in-service bridge. This framework promises significant advancements in assessing aging bridge infrastructure with higher accuracy and efficiency compared to analytical or semi-analytical approaches.

Data‐driven target localization using adaptive radar processing and convolutional neural networks

AbstractLeveraging the advanced functionalities of modern radio frequency (RF) modeling and simulation tools, specifically designed for adaptive radar processing applications, this paper presents a data‐driven approach to improve accuracy in radar target localization post adaptive radar detection. To this end, we generate a large number of radar returns by randomly placing targets of variable strengths in a predefined area, using RFView®, a high‐fidelity, site‐specific, RF modeling & simulation tool. We produce heatmap tensors from the radar returns, in range, azimuth [and Doppler], of the normalized adaptive matched filter (NAMF) test statistic. We then train a regression convolutional neural network (CNN) to estimate target locations from these heatmap tensors, and we compare the target localization accuracy of this approach with that of peak‐finding and local search methods. This empirical study shows that our regression CNN achieves a considerable improvement in target location estimation accuracy. The regression CNN offers significant gains and reasonable accuracy even at signal‐to‐clutter‐plus‐noise ratio (SCNR) regimes that are close to the breakdown threshold SCNR of the NAMF. We also study the robustness of our trained CNN to mismatches in the radar data, where the CNN is tested on heatmap tensors collected from areas that it was not trained on. We show that our CNN can be made robust to mismatches in the radar data through few‐shot learning, using a relatively small number of new training samples.

Sea Surface Wind Speed Estimation From the Combination of Satellite Scatterometer and Radiometer Parameters

AbstractSatellite research on global sea surface winds is crucial for understanding and monitoring the dynamics of earth's oceans, providing valuable insights into weather patterns, climate changes, and oceanographic processes. This study investigates the fusion of active (microwave scatterometer) and passive (microwave radiometer) satellite data using Machine learning (ML) for sea surface wind speed estimation. Employing Random Forest Regression (RF), Convolutional Neural Network Regression (CNN), and Multiple Linear Regression (MLR). Evaluation against reference data sets (Advanced Scatterometer, ERA5, Cross‐Calibrated Multi‐Platform (CCMP), buoy wind speeds) highlights the robustness of the proposed models. The research findings indicate that both RF and CNN exhibit superior accuracy in the active, passive, and joint active‐passive models compared to the simplistic MLR model. The joint models of the three regression methods outperform the individual active or passive models. The root mean square deviation (RMSD) accuracy of RF and CNN joint models, when compared with ASCAT, is in the order of 0.7 m/s, and when compared with buoys, the RMSD accuracy is around 1.1 m/s. The ML models, especially RF and CNN, demonstrate superior performance, providing accurate and reliable estimations crucial for meteorological and oceanographic applications. These findings underscore the potential operational use of ML techniques in enhancing remote sensing applications.

Distributional loss for convolutional neural network regression and application to parameter estimation in satellite navigation signals

Convolutional Neural Network (CNN) have been widely used in image classification. Over the years, they have also benefited from various enhancements and they are now considered state-of-the-art techniques for image-like data. However, when they are used for regression to estimate some function value from images, few recommendations are available to construct robust CNN regressor models. In this study, a robustness enforcing mechanism is proposed for CNN regression models. It combines convolutional neural layers to extract high level features representations from images with a soft labelling technique that helps generalization performance. More specifically, as the deep regression task is challenging, the idea is to account for some uncertainty in the targets that are seen as distributions around their mean. Building from earlier work (Imani and White, 2018), a specific histogram loss function based on the Kullback-Leibler (KL) divergence is applied during training. The prior distributions are selected according to the physical characteristics of the parameters to estimate. To assess and illustrate the technique, the model is applied to Global Navigation Satellite System (GNSS) multipath estimation where multipath signal parameters have to be estimated from correlator output images from the I and Q channels. The multipath signal delay, magnitude, Doppler shift frequency and phase parameters are estimated from synthetically generated datasets of satellite signals. Experiments are conducted under various receiving conditions and various input images resolutions to test the estimation performances quality and robustness. The results show that the proposed soft labelling CNN technique using distributional loss outperforms classical CNN regression under all conditions. Furthermore, the extra learning performance achieved by the model allows the reduction of input image resolution from 80x80 down to 40x40 or sometimes 20x20.

AI-powered sensor fault detection for cost-effective smart greenhouses

Sensor networks in greenhouses play a pivotal role in controlling the stability of environmental and chemical factors. The Internet of Things (IoT) has been widely adopted for monitoring various sensor networks in greenhouses. In the present study, an IoT platform for remote monitoring of greenhouse environment is designed. The platform consists of a sensor node, a sink node, and an edge server. The sensor node measures indoor and outdoor humidity and temperature, interior CO(g), and interior luminosity and transmits data to the sink node, where it is timestamped. The sink node is connected to an edge server through the Mosquitto MQTT broker and the data is subsequently transferred to a MongoDB Cloud infrastructure, where the data of each variable is stored in proper formats. In the second part of the study, four 1D convolutional neural networks (CNNs) were developed for data prediction of each sensor to provide fault-tolerance in the system. The first three models, for predicting inner humidity and temperature, outdoor temperature, and outdoor humidity, directly predict the actual data of the faulty sensor based on regression analysis. The last model is designed for predicting CO and luminosity and performs data classification for faulty sensors. The models provide a high level of precision in their predictions. The RMSE for interior temperature and humidity and exterior temperature and humidity are 0.86 ℃, 3.47%, 0.682 ℃, and 2.74%, respectively. Additionally, the accuracy for luminosity and CO classification are 89.70% and 83.43%. Comparison of 1D CNN, decision tree, and linear regression model revealed that the machine learning models considerably outperform linear regression model, and they can better capture the nonlinear correlations among the variables. Furthermore, the predictive outcomes of the models were consistent across different weather conditions. The proposed methodology can be used to induce tolerance against faulty reads at sensor level in greenhouse sensor networks, independent of time and the data gathered by the faulty sensors. The study indicates that exploiting cloud resources for promoting the use of complex AI models on IoT platforms can provide a suitable solution for real-time monitoring and fault control of greenhouse environmental factors.

Convolutional neural network regression for low-cost microalgal density estimation

Density of microalgae is critical information for production of algae in a closed cultivation system since it can be used to optimally control their growth rate, biomass concentration and quality of the products. Given advancement in image processing techniques and thanks to low-cost camera sensors, image based methods are increasingly widely utilized to indirectly estimate the density. Advantages of the image based techniques include being less invasive and more nondestructive and biosecured. However, most of the existing techniques rely on averaging all pixels of a microalgae image, which may eliminate crucial information of their spatial correlation. Therefore, in this work we propose to exploit a convolutional neural network (CNN) to efficiently extract information from the microalgae images, which are then employed to regress the density. Interestingly, the proposed deep CNN regression model accepts the whole color image as its input while the density is calculated in the output. The proposed CNN regression architecture was validated in real-world experiments where the microalgal strain Chlorella vulgaris was cultured and their images were captured by our low-cost camera sensor system. The obtained results demonstrate that the averaged estimation accuracy of the proposed model is 99.45% (± 0.68%) while the R2 value between the density predictions and the ground truths is 0.9997, which is highly accurate and practical.

Precision classification and quantitative analysis of bacteria biomarkers via surface-enhanced Raman spectroscopy and machine learning

The SERS spectra of six bacterial biomarkers, 2,3-DHBA, 2,5-DHBA, Pyocyanin, lipoteichoic acid (LTA), Enterobactin, and β-carotene, of various concentrations, were obtained from silver nanorod array substrates, and the spectral peaks and the corresponding vibrational modes were identified to classify different spectra. The spectral variations in three different concentration regions due to various reasons have imposed a challenge to use classic calibration curve methods to quantify the concentration of biomarkers. Depending on baseline removal strategy, i.e., local or global baseline removal, the calibration curve differed significantly. With the aid of convolutional neural network (CNN), a two-step process was established to classify and quantify biomarker solutions based on SERS spectra: using a specific CNN model, a remarkable differentiation and classification accuracy of 99.99 % for all six biomarkers regardless of the concentration can be achieved. After classification, six regression CNN models were established to predict the concentration of biomarkers, with coefficient of determination R2 > 0.97 and mean absolute error (MAE) < 0.27. The feature of important calculations indicates the high classification and quantification accuracies were due to the intrinsic spectral features in SERS spectra. This study showcases the synergistic potential of SERS and advanced machine learning algorithms and holds significant promise for bacterial infection diagnostics.

SonoMyoNet: A Convolutional Neural Network for Predicting Isometric Force from Highly Sparse Ultrasound Images.

Ultrasound imaging or sonomyography has been found to be a robust modality for measuring muscle activity due to its ability to image deep-seated muscles directly while providing superior spatiotemporal specificity compared to surface electromyography-based techniques. Quantifying the morphological changes during muscle activity involves computationally expensive approaches for tracking muscle anatomical structures or extracting features from brightness-mode (B-mode) images and amplitude-mode (A-mode) signals. This paper uses an offline regression convolutional neural network (CNN) called SonoMyoNet to estimate continuous isometric force from sparse ultrasound scanlines. SonoMyoNet learns features from a few equispaced scanlines selected from B-mode images and utilizes the learned features to estimate continuous isometric force accurately. The performance of SonoMyoNet was evaluated by varying the number of scanlines to simulate the placement of multiple single-element ultrasound transducers in a wearable system. Results showed that SonoMyoNet could accurately predict isometric force with just four scanlines and is immune to speckle noise and shifts in the scanline location. Thus, the proposed network reduces the computational load involved in feature tracking algorithms and estimates muscle force from the global features of sparse ultrasound images.

Deep learned features selection algorithm: Removal operation of anomaly feature maps (RO-AFM)

Class/Regression Activation Maps (CAMs/RAMs; AMs) are often embedded into Convolutional Neural Networks (CNNs) for checking activated regions on input images at estimation. CNNs sometime generate unreliable AMs, such as activated regions, are inappropriate. Because AM is calculated by stacking many feature maps generated by the final convolutional layer, when there are Anomaly Feature Maps (AFMs), unreliable AMs can be generated. For example, suppose we have a CNN that evaluates the heart. In this case, the feature maps that focus on regions unrelated to the heart (e.g., shoulders and esophagus) are AFMs. Additionally, we have a hypothesis that the estimation accuracy of CNNs is increased by removing AFMs. However, methods for automatically detecting and removing AFMs have not been sufficiently studied in previous research to improve the performance of CNNs. Therefore, we propose a method named “Removal Operation of Anomaly Feature Maps (RO-AFMs)” to automatically detect and remove AFMs. When applying an RO-AFM to the Global Average Pooling (GAP) feature vectors of a CNN, dimensions of the GAP vector are reduced. Therefore, an RO-AFM is regarded as a deep-feature selection algorithm. From the results of adopting an RO-AFM to a Regression CNN (R-CNN) for estimating pulmonary artery wedge pressure, which is one of the measurement score for representing cardiac anomaly state, improved reliability of AM and estimation accuracy were verified. A comparison of RO-AFM and the existing methods, i.e., Lasso and the Feature Selection Layer (FSL), indicated that RO-AFM performed slightly better on the estimation accuracy. The computation time required for RO-AFM to evaluate all features was 1.833 s on average, confirming that RO-AFM is a lightweight process. Therefore, RO-AFM is useful for constructing a medical CNN that emphasizes explainability (e.g., CNNs for estimating the risk of a disease or a test value from chest X-ray or computed tomography images).

Physics-Informed Approximation of Internal Thermal History for Surface Deformation Predictions in Wire Arc Directed Energy Deposition

Abstract This work presents a physics-informed fusion methodology for deformation detection using multi-sensor thermal data. A challenge with additive manufacturing (AM) is that abnormalities commonly occur due to rapid changes in the thermal gradient. Different non-destructive in-situ thermal sensors capture parts of the thermal history but are limited by the visible temperature spectrum and sensor field of view of the fabrication process. Various sensors mitigate problems with the loss of thermal history information; however, it brings forth challenges with integrating different data streams and the need to interpolate the internal thermal history. This study develops a thermal data-informed heat flux methodology that fills the gap in fusing numerical temperature approximation with data-driven knowledge of the surface of additive manufactured components. First, this study fuses infrared (IR) thermal data complexities during the AM process with the Goldak double ellipsoidal heat flux to model the energy input into the component. Second, a thermal physics-informed model input (PIMI) is created with thermal data-informed heat flux to capture internal thermal history. Lastly, a regression convolutional neural network (CNN) captures the relationship between the three-dimensional thermal gradient and the resulting surface deformation. The rapid thermal gradient formation and identification of deformation is a key step toward using thermal history data and machine learning to improve quality control in AM. The proposed surface deformation detection model achieved an mean squared error of 1.14 mm and an R2 of 0.89 in the case study when fabricating thin-walled structures.

Deep learning with plasma plume image sequences for anomaly detection and prediction of growth kinetics during pulsed laser deposition

Materials synthesis platforms that are designed for autonomous experimentation are capable of collecting multimodal diagnostic data that can be utilized for feedback to optimize material properties. Pulsed laser deposition (PLD) is emerging as a viable autonomous synthesis tool, and so the need arises to develop machine learning (ML) techniques that are capable of extracting information from in situ diagnostics. Here, we demonstrate that intensified-CCD image sequences of the plasma plume generated during PLD can be used for anomaly detection and the prediction of thin film growth kinetics. We develop multi-output (2 + 1)D convolutional neural network regression models that extract deep features from plume dynamics that not only correlate with the measured chamber pressure and incident laser energy, but more importantly, predict parameters of an auto-catalytic film growth model derived from in situ laser reflectivity experiments. Our results demonstrate how ML with in situ plume diagnostics data in PLD can be utilized to maintain deposition conditions in an optimal regime. Further, the predictive capabilities of plume dynamics on the kinetics of film growth or other film properties prior to deposition provides a means for rapid pre-screening of growth conditions for the non-expert, which promises to accelerate materials optimization with PLD.

Estimation of non-uniform motion blur using a patch-based regression convolutional neural network

The non-uniform blur of atmospheric turbulence can be modeled as a superposition of linear motion blur kernels at a patch level. We propose a regression convolutional neural network (CNN) to predict angle and length of a linear motion blur kernel for varying sized patches. We analyze the robustness of the network for different patch sizes and the performance of the network in regions where the characteristics of the blur are transitioning. Alternating patch sizes per epoch in training, we find coefficient of determination scores across a range of patch sizes of R2&gt;0.78 for length and R2&gt;0.94 for angle prediction. We find that blur predictions in regions overlapping two blur characteristics transition between the two characteristics as overlap changes. These results validate the use of such a network for prediction of non-uniform blur characteristics at a patch level.

Individual tree detection in large-scale urban environments using high-resolution multispectral imagery

Systematic maps of urban forests are useful for regional planners and ecologists to understand the spatial distribution of trees in cities. However, manually-created urban forest inventories are expensive and time-consuming to create and typically do not provide coverage of private land. Toward the goal of automating urban forest inventory through machine learning techniques, we performed a comparative study of methods for automatically detecting and localizing trees in multispectral aerial imagery of urban environments, and introduce a novel method based on convolutional neural network regression. Our evaluation is supported by a new dataset of over 1,500 images and almost 100,000 tree annotations, covering eight cities, six climate zones, and three image capture years. Our method outperforms previous methods, achieving 73.6% precision and 73.3% recall when trained and tested in Southern California, and 76.5% precision 72.0% recall when trained and tested across the entire state. To demonstrate the scalability of the technique, we produced the first map of trees across the entire urban forest of California. The map we produced provides important data for the planning and management of California’s urban forest, and establishes a proven methodology for potentially producing similar maps nationally and globally in the future.

Improved CNN-based model for shape parameter determination of sand particles considering imbalanced dataset

This study presents an artificial intelligence prediction model for determining the roundness and sphericity of sand particles using a convolutional neural network (CNN) regression model, which combines a CNN and a regression model. The dataset used for the model development consisted of 27,000 binary images of coarse grains with various shapes. To develop the prediction model, the validated algorithms, ResNet101 and ResNet152, were utilized and modified into a CNN regression model to predict the shape parameters. The performance of the developed model was evaluated using 3,000 randomly generated binary images, and the mean squared error and root mean squared error were calculated as evaluation metrics. The evaluation results indicated that the model effectively predicted the sphericity and roundness of sand particles, with a relatively low prediction performance observed for roundness.

In-Season Cotton Yield Prediction with Scale-Aware Convolutional Neural Network Models and Unmanned Aerial Vehicle RGB Imagery.

In the pursuit of sustainable agriculture, efficient water management remains crucial, with growers relying on advanced techniques for informed decision-making. Cotton yield prediction, a critical aspect of agricultural planning, benefits from cutting-edge technologies. However, traditional methods often struggle to capture the nuanced complexities of crop health and growth. This study introduces a novel approach to cotton yield prediction, leveraging the synergy between Unmanned Aerial Vehicles (UAVs) and scale-aware convolutional neural networks (CNNs). The proposed model seeks to harness the spatiotemporal dynamics inherent in high-resolution UAV imagery to improve the accuracy of the cotton yield prediction. The CNN component adeptly extracts spatial features from UAV-derived imagery, capturing intricate details related to crop health and growth, modeling temporal dependencies, and facilitating the recognition of trends and patterns over time. Research experiments were carried out in a cotton field at the USDA-ARS Cropping Systems Research Laboratory (CSRL) in Lubbock, Texas, with three replications evaluating four irrigation treatments (rainfed, full irrigation, percent deficit of full irrigation, and time delay of full irrigation) on cotton yield. The prediction revealed that the proposed CNN regression models outperformed conventional CNN models, such as AlexNet, CNN-3D, CNN-LSTM, ResNet. The proposed CNN model showed state-of-art performance at different image scales, with the R2 exceeding 0.9. At the cotton row level, the mean absolute error (MAE) and mean absolute percentage error (MAPE) were 3.08 pounds per row and 7.76%, respectively. At the cotton grid level, the MAE and MAPE were 0.05 pounds and 10%, respectively. This shows the proposed model's adaptability to the dynamic interplay between spatial and temporal factors that affect cotton yield. The authors conclude that integrating UAV-derived imagery and CNN regression models is a potent strategy for advancing precision agriculture, providing growers with a powerful tool to optimize cultivation practices and enhance overall cotton productivity.