Early fruit-yield forecasting plays an important role in productive and sustainable management of apple orchards. This paper presents a method combining image processing with support vector machine (SVM) technology to build a prediction model for early season apple tree yield estimation. Sixty ‘Gala’ apple trees were randomly selected for study. Initially, tree canopy images were captured in natural light just after June drop when the fruit color was green. Apples in the canopy image were identified with the condition Cb≤100 and Cr≥120 obtained by analyzing the distribution map of color component values in YCbCr color space, in which Y was the luminance component, Cb and Cr were the blue-difference and red-difference chroma components. By the same method, the condition Cr≤125 was used to segment foliage from canopy image with fruit removed. Five characteristics were extracted from the canopy image: fruit total area, total number of fruit, proportion of fruit total area to foliage area, proportion of total fruit area shaded by leaves to total fruit area, and proportion of total fruit numbers shaded by leaves to total fruit number. Finally, the SVM method was employed to build a yield estimation model with these five characteristics as input parameters and the actual yield as output. A randomized sample set containing 50 trees was used to train the model, yielding a coefficient of determination (R2) of 0.7242, a root mean square error (RMSE) of 1.71kg, a mean absolute percentage error (MAPE) of 9% and an average prediction error (MFE) of 0.21. Using 15 independent samples, the model was validated, yielding a RMSE of 2.45kg and a MAPE of 13%. The proposed model showed significant potential for early apple yield prediction of individual trees with potential application to other fruit crops.