我正在嘗試僅使用 NumPy 來實(shí)現(xiàn)圖像卷積代碼，類似于cv2.filter2D(...)的做法。import numpy as npimport time# kernalH = np.array([[0,1,0],[1,-4,1],[0,1,0]])# imageSEED = 23img = np.random.RandomState(SEED).randint(10, size=(4, 4))# shapesHi, Wi = img.shapeHk, Wk = H.shapehk = Hk//2wk = Wk//2# paddingnew_img = np.pad(img, (hk, wk), 'constant', constant_values=0)pHi, pWi = new_img.shapeprint('img: ')print(new_img)print('kernal: ')print(H)print('\n')# image convolution################################################### Method 1st = time.time()out = np.zeros((Hi, Wi))for i in range(hk, pHi-hk):    for j in range(wk, pWi-wk):        batch = new_img[i-hk:i+hk+1, j-wk:j+wk+1]        out[i-hk][j-wk] = np.sum(batch*H)print('out: ')print(out)print("[*] process time : %f" % (time.time()- st))print('\n')################################################### Method 2st = time.time()Hi, Wi = img.shapeout = np.zeros((Hi, Wi))H_1d = H.ravel()hl = len(H_1d)for i in range(Wi):    img_slice = new_img[:, i:Hk+i]    for j in range(Hi):        out[j][i] = np.sum(img_slice[j:j+Hk,:]*H)    il = len(img_slice)    h1 = 0    h2 = hl        while h2 <= il:        index = h1//Hk        # print(index)        out[index][i] = np.sum(img_slice[h1:h2]*H_1d)        h1 = h1 + Hk        h2 = h2 + Hkprint('out: ')print(out)print("[*] process time : %f" % (time.time()- st))print('\n')##################################################它有效，但我需要更快、更有效的實(shí)施。為了使算法更快，我嘗試在此處查看圖像的卷積過程中找到模式，但仍然無法弄清楚應(yīng)該做什么。有人可以幫助我改進(jìn)當(dāng)前的代碼，使其更快，僅使用 NumPy 嗎？