Abstract: Tires are the only components that come into contact with the ground and continuously interact with it during the operation of agricultural vehicles. At the same time, tires are also the core of carrying and driving agricultural machinery such as tractors, and the working state of tires directly affects various performance of agricultural vehicles. The typical characteristics of agricultural tires include large load fluctuations, special pattern shapes, harsh working environments, and significant tire body vibrations, which make it difficult to accurately obtain tire vertical loads in practical operations. However, vertical load has a significant impact on the performance of agricultural machinery and is a key factor in evaluating and optimizing the efficiency and stability of agricultural machinery operations. An off-road tire vertical load estimation method based on bidirectional strain of wheel rims is proposed to address the difficulties in obtaining vertical loads for agricultural tires and the low estimation accuracy of traditional models. Taking off-road tire as the research object, the radial and tangential strains of the wheel rim under vertical load were analyzed, and a wheel rim strain acquisition system based on high-precision pin axis sensors was designed. Multiple typical working condition tests were conducted using a tire drum test bench, and the strain curve of the wheel rim during tire rolling was obtained. We conducted data denoising, period partitioning, data filtering, and period feature extraction on the corresponding strain curve, and established a dataset of tangential and radial strains of the wheel rim during tire rolling process. A tire vertical load estimation model was constructed using a Self-Attention Convolutional Neural Network (AT-CNN), which takes periodic tangential strain curve and periodic radial strain curve as inputs. The model consists of two single load prediction networks and a coupled network. Two single load prediction networks input periodic radial strain curves and periodic tangential curves, respectively. The coupled network takes the corresponding two periodic radial and tangential curves as inputs, and they both output predicted values of vertical loads. The estimation results show that the Mean Absolute Error (MAE) of the AT-CNN estimation model is 95.68N, and the Root Mean Square Error (RMSE) is 100.54N. The MAE of the traditional deep neural network (DNN) model is 557.35N, and the RMSE is 588.07N. Compared with the DNN network, the MAE of AT-CNN is reduced by 82.83%, and the RMSE is reduced by 82.90%. In order to understand the working principle and feature extraction process of load prediction networks, the t-distributed stochastic neighbor embedding (t-SNE) method is applied to reduce the dimensionality of features in the network and visualize them. The visualization results show that after the input data is processed by the AT-CNN, the similarity of features with the same strain becomes closer and the classification features become more obvious. Analyze the generalization ability of AT-CNN through a 10-fold cross test. The experimental results show that in ten experiments, the average MAE of DNN is 647.32N, and the average RMSE is 650.16N. In ten experiments, the average MAE of AT-CNN is 98.73N and the average RMSE is 102.33N. Compared with the DNN model, the AT-CNN model has higher prediction accuracy, and the MAE and RMSE distributions are more concentrated, indicating that the AT-CNN model has higher prediction accuracy and generalization ability. Verify the ability of key modules such as self-attention module and feature fusion module to improve the overall performance of the AT-CNN through ablation experiments. The experimental results show that the self-attention module and feature fusion module significantly improved the estimation accuracy of the network. Research has shown that the proposed off-road tire vertical load estimation method based on bidirectional strain of the wheel rim achieves real-time and accurate estimation of the vertical load of off-road tires.