This study evaluated the ability of the YOLOv8 network, the VGG algorithm, and the ResNet architecture to predict carcass traits through the recognition of real-time ultrasonographic images of sheep carcasses. A total of 263 ultrasonographic images of the aforementioned muscle were collected from meat-type sheep. The measurements of loineye area (LEA), fat thickness (SFT), length (LEL), and depth (LED) were recorded in cm², mm, cm, and cm; respectively, using ultrasound performed by a trained professional, while the Ratio was expressed in absolute values. The collected images were uploaded to the tool to delimit the measurements within the image and were subsequently compiled for processing and extraction of LEA, SFT, LEL, and LED measurements using the software ImageJ. The image dataset was divided into original images and images containing the SFT measurement (segmentation masks) to identify the region of interest in standardized images (224 × 224 × 3 pixels). During the segmentation stage, the YOLOv8 network was used to process and adjust the images, while the VGG and ResNet algorithms were used to validate and present the results. The dataset was split into training (80%), validation (10%), and testing (10%). Precision, mAP50, and recall metrics were used to evaluate the similarity between the two image sets. To estimate LEA, SFT, LEL, and LED from neural network descriptors, that is, the features extracted from an image or feature vectors derived from segmented images, three regressors were used: AdaBoost Regressor (ABR), Random Forest Regressor (RFR), and Gradient Boosting Regressor (GBR). To evaluate the quality of the regression fit, the mean squared residue (MSR), mean absolute error (MAE), and coefficient of determination (R²) were calculated. To compare the actual and predicted means of the measurements, analysis of variance and Student’s t-test were performed at a 5% significance level. Among the regressors tested, the RFR showed the best performance with the evaluated algorithms, demonstrating the robustness of this regressor. The similarity between the expert-labeled images and those generated by the algorithm was also assessed, showing agreement between the two sets. The means predicted by the tool and those obtained by the specialist did not differ significantly. Predicted values showed lower variance than the measured values, indicating greater precision in the automated estimates produced by the tool. Therefore, the automatic measurement of these traits using the proposed tool based on ultrasonographic images is promising. In addition, the method allows the analysis of a large number of images in a short period of time, enabling faster feedback of information to sheep producers. However, due to the relatively low fit of the equations to the data, additional image collection in field conditions is necessary to strengthen sheep phenotyping.